Health AI Benchmarks

Sources

29 documents, each with the results it supports.

System cards

  1. Addendum to GPT-6 Astra System Card: GPT-6.1 Terra System card from OpenAI, published 26 October 2026, checked 2 November 2026. First-party.

    2 results from this document
    • GPT-6.1 TerraonHealthBench Professional62.7

      OpenAI; length-adjusted; 62.7 (64.2, 2531) (adjusted, raw, mean response characters); addendum Table 5 refers to the GPT-6 Astra card for method

      “EVALUATION GPT-5.5 GPT-5.6-SOL GPT-5.6-TERRA GPT-5.6-LUNA GPT-6 ASTRA GPT-6-SOL GPT-6-LUNA GPT-6.1-SOL GPT-6.1-TERRA HealthBench Professional length-adjusted 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389) 64.7 (68.2, 3185) 60.8 (59.5, 1573) 60.8 (61.2, 2119) 64.2 (67.2, 3038) 62.7 (64.2, 2531)”

      Section 5.1, Table 5, HealthBench Professional length-adjusted, GPT-6.1 Terra column

    • GPT-6.1 TerraonHealthBench55.1

      OpenAI; length-adjusted; 55.1 (51.1, 1331) (adjusted, raw, mean response characters); addendum Table 5 refers to the GPT-6 Astra card for method

      “EVALUATION GPT-5.5 GPT-5.6-SOL GPT-5.6-TERRA GPT-5.6-LUNA GPT-6 ASTRA GPT-6-SOL GPT-6-LUNA GPT-6.1-SOL GPT-6.1-TERRA HealthBench length-adjusted 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930) 58.3 (56.9, 1760) 53.2 (47.1, 977) 54.5 (50.0, 1255) 58.5 (56.7, 1701) 55.1 (51.1, 1331)”

      Section 5.1, Table 5, HealthBench length-adjusted, GPT-6.1 Terra column

  2. Claude Haiku 5.5 System Card System card from Anthropic, published 7 October 2026, checked 12 October 2026. First-party.

    5 results from this document
    • GPT-6 Luna (Anthropic run)onHealthBench Professional65.5%

      Run by Anthropic through OpenAI's public API; max effort; Opus 4.8 grader; length-adjusted; not OpenAI's published figure

      “HealthBench Professional panel: 65.5% (GPT-6 Luna (OpenAI)); caption: GPT-6 Luna's scores come from our own runs through OpenAI's public API and are not OpenAI's published figures.”

      p. 132, Figure 8.11.1.A, HealthBench Professional panel, GPT-6 Luna (OpenAI) diamond label (visually read printed label); Section 8.11.1 text on pp. 131-132

    • Claude Haiku 5.5onHealthBench Professional64.8%

      Anthropic; max effort; Opus 4.8 grader; safety classifiers enabled; length-adjusted; raw 71.0%; five runs (October 1 and 2); thirteen repeat max runs on October 4 averaged 64.2%

      “Max effort scored highest, at 64.8% (71.0% raw); thirteen repeat runs at max on October 4, the day of the lower levels, averaged 64.2%.”

      p. 133, Section 8.11.3; same value in p. 111, Table 8.1.A, row HealthBench Professional (Length-adjusted), column Haiku 5.5 (printed 64.8) and p. 132, Figure 8.11.1.A

    • Claude Haiku 4.5onHealthBench Professional32.2

      Anthropic; extended thinking (Haiku 4.5 has no effort setting); Opus 4.8 grader; safety classifiers enabled; length-adjusted; measured for the Haiku 5.5 card

      “Evaluation | Haiku 5.5 | Haiku 4.5 | Sonnet 5.5 | GPT-6 Luna HealthBench Professional (Length-adjusted) | 64.8 | 32.2 | 69.2 | –”

      p. 111, Table 8.1.A, row HealthBench Professional (Length-adjusted), column Haiku 4.5

    • Claude Haiku 5.5onHealthBench61.6%

      Anthropic; max effort; Opus 4.8 grader; safety classifiers enabled; length-adjusted; raw 64.2%; mean of five runs made on October 2

      “On HealthBench (one run per level below max), Haiku 5.5 scored 59.9% at low, 60.2% at medium, 60.3% at high, 61.1% at xhigh and 61.6% at max, a rise of 1.8 points against 6.9 on HealthBench Professional”

      p. 133, Section 8.11.3; p. 132, Figure 8.11.1.A HealthBench panel labels 'Haiku 5.5 61.6%' (length-adjusted, solid) and 'raw 64.2%'

    • GPT-6 Luna (Anthropic run)onHealthBench56.3%

      Run by Anthropic through OpenAI's public API; max effort; Opus 4.8 grader; length-adjusted; not OpenAI's published figure

      “HealthBench panel: 56.3% (GPT-6 Luna (OpenAI)); caption: GPT-6 Luna's scores come from our own runs through OpenAI's public API and are not OpenAI's published figures.”

      p. 132, Figure 8.11.1.A, HealthBench panel, GPT-6 Luna (OpenAI) diamond label (visually read printed label); Section 8.11.1 text on pp. 131-132

  3. Addendum to GPT-6 Astra System Card: GPT-6.1 Sol System card from OpenAI, published 29 September 2026, checked 6 October 2026. First-party.

    2 results from this document
    • GPT-6.1 SolonHealthBench Professional64.2

      OpenAI; length-adjusted; 64.2 (67.2, 3038) (adjusted, raw, mean response characters); addendum Table 7 refers to the GPT-6 Astra card for method

      “EVALUATION | GPT-5.5 | GPT-5.6 SOL | GPT-5.6 TERRA | GPT-5.6 LUNA | GPT-6 ASTRA | GPT-6 SOL | GPT-6 LUNA | GPT-6.1 SOL HealthBench Professional length-adjusted | 51.8 (57.2, 3818) | 60.5 (64.1, 3228) | 57.7 (62.4, 3618) | 55.7 (59.8, 3389) | 64.7 (68.2, 3185) | 60.8 (59.5, 1573) | 60.8 (61.2, 2119) | 64.2 (67.2, 3038)”

      Section 5.1, Table 7, HealthBench Professional length-adjusted, GPT-6.1 Sol column

    • GPT-6.1 SolonHealthBench58.5

      OpenAI; length-adjusted; 58.5 (56.7, 1701) (adjusted, raw, mean response characters); addendum Table 7 refers to the GPT-6 Astra card for method

      “EVALUATION | GPT-5.5 | GPT-5.6 SOL | GPT-5.6 TERRA | GPT-5.6 LUNA | GPT-6 ASTRA | GPT-6 SOL | GPT-6 LUNA | GPT-6.1 SOL HealthBench length-adjusted | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930) | 58.3 (56.9, 1760) | 53.2 (47.1, 977) | 54.5 (50.0, 1255) | 58.5 (56.7, 1701)”

      Section 5.1, Table 7, HealthBench length-adjusted, GPT-6.1 Sol column

  4. Claude Sonnet 5.5 System Card System card from Anthropic, published 28 September 2026. First-party.

    3 results from this document
    • GPT-6 Astra (Anthropic run)onHealthBench Professional70.3%

      Anthropic public-API reproduction; max effort; no system prompt; Opus 4.8 grader; length-adjusted; raw 74.0%

      “On HealthBench Professional at max effort, length adjustment changes the ranking. After: GPT-6 Astra (70.3%) > Claude Sonnet 5.5 (69.2%) > Claude Opus 5.5 (65.6%) > Claude Fable 5.1 (62.1%).”

      p. 138, paragraph above Figure 8.15.B, HealthBench Professional after length adjustment.

    • Claude Sonnet 5.5onHealthBench Professional69.2%

      Anthropic; max effort; Opus 4.8 grader; safety classifiers enabled; length-adjusted; raw 77.1%

      “On HealthBench Professional at max effort, length adjustment changes the ranking. After: GPT-6 Astra (70.3%) > Claude Sonnet 5.5 (69.2%) > Claude Opus 5.5 (65.6%) > Claude Fable 5.1 (62.1%).”

      p. 138, Section 8.15.2 and paragraph above Figure 8.15.B.

    • Claude Sonnet 5.5onHealthBench65.4%

      Anthropic; max effort; Opus 4.8 grader; safety classifiers enabled; length-adjusted; raw 69.4%

      “Sonnet 5.5 | HealthBench | max 65.4%”

      p. 138, Figure 8.15.B, left panel HealthBench; orange Sonnet 5.5 max-effort label (visually read printed labels).

  5. Claude Opus 5.5 System Card System card from Anthropic, published 22 September 2026, checked 1 November 2026. First-party.

    3 results from this document
    • Claude Opus 5.5onHealthBench Professional65.6%

      Anthropic; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt; safety classifiers and refusal fallback to Opus 5; length-adjusted; raw 77.1%

      “8.15.2 HealthBench Professional results length adjustment, which penalizes verbose model responses, Claude Opus 5.5 achieved a score of 65.6%.”

      pp. 213–214, Section 8.15.2, paragraph and Figure 8.15.2.A.

    • Claude Opus 5.5onHealthBench68.1%

      raw/unadjusted, Anthropic system-card protocol (adaptive max effort, Opus 4.8 grader, five trials, no tools or custom system prompt; safety classifiers and refusal fallback to Opus 5); length-adjusted 60.6

      “On HealthBench, Claude Opus 5.5 achieved a raw score of 68.1%, ahead of Claude Fable 5.1 at 66.7%, Claude Sonnet 5 at 59.2%, and Claude Opus 5 at 67.1%. After length adjustment, which penalizes verbose model responses, Claude Opus 5.5 achieved a score of 60.6%.”

      p. 213, section 8.15.1 HealthBench results; Figure 8.15.1.A

    • Claude Opus 5.5onHealthBench60.6%

      Anthropic; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt; safety classifiers and refusal fallback to Opus 5; length-adjusted; raw 68.1%

      “8.15.1 HealthBench results After length adjustment, which penalizes verbose model responses, Claude Opus 5.5 achieved a score of 60.6%.”

      p. 213, Section 8.15.1, paragraph and Figure 8.15.1.A.

  6. GPT-6 Astra System Card System card from OpenAI, published 3 September 2026, checked 8 September 2026. First-party.

    2 results from this document
    • GPT-6 AstraonHealthBench Professional63.4

      length-adjusted, max reasoning effort (68.2 unadjusted, 3,185 mean response chars, as corrected by OpenAI on September 22; first printed as 69.5 and 4,097); GPT-6 Astra system card Table 6, column 'gpt-6 Astra'.

      This document disagrees with the primary source.

      “Astra has a length-adjusted HealthBench Professional score of 63.4 (+2.9 relative to GPT-5.6 Sol), HealthBench score of 58.1 (+1.1), HealthBench Hard score of 36.3 (+3.2), and HealthBench Consensus score of 95.8 (+0.3).”

      p. 19, sec. 6.1 (Table 6 cell, column 'gpt-6 Astra': 63.4 (69.5, 4097))

    • GPT-6 AstraonHealthBench58.1

      length-adjusted, max reasoning effort (59.7 unadjusted, 2,258 mean response chars); GPT-6 Astra system card Table 6, column 'gpt-6 Astra'.

      This document disagrees with the primary source.

      “Astra has a length-adjusted HealthBench Professional score of 63.4 (+2.9 relative to GPT-5.6 Sol), HealthBench score of 58.1 (+1.1), HealthBench Hard score of 36.3 (+3.2), and HealthBench Consensus score of 95.8 (+0.3).”

      p. 19, sec. 6.1 (Table 6 cell, column 'gpt-6 Astra': 58.1 (59.7, 2258))

  7. GPT-6 Astra System Card - HealthBench (Deployment Safety Hub) System card from OpenAI, published 3 September 2026, checked 8 September 2026. First-party.

    6 results from this document
    • GPT-6 AstraonHealthBench Professional64.7

      length-adjusted, max reasoning effort (68.2 unadjusted, 3,185 mean response chars, as corrected by OpenAI on September 22; first printed as 69.5 and 4,097); GPT-6 Astra system card Table 6, column 'gpt-6 Astra'.

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench Professional length-adjusted | 51.8 (57.2, 3818) | 60.5 (64.1, 3228) | 57.7 (62.4, 3618) | 55.7 (59.8, 3389) | 64.7 (68.2, 3185) | 60.8 (59.5, 1573) | 60.8 (61.2, 2119)”

      Section 11.4.1, Table 29, HealthBench Professional length-adjusted, GPT-6 Astra column; September 22 correction.

    • GPT-6 LunaonHealthBench Professional60.8

      OpenAI; length-adjusted; maximum reasoning effort; 60.8 (61.2, 2119) (adjusted, raw, mean response characters)

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench Professional length-adjusted | 51.8 (57.2, 3818) | 60.5 (64.1, 3228) | 57.7 (62.4, 3618) | 55.7 (59.8, 3389) | 64.7 (68.2, 3185) | 60.8 (59.5, 1573) | 60.8 (61.2, 2119)”

      Section 11.4.1, Table 29, HealthBench Professional length-adjusted, GPT-6 Luna column; September 22 correction.

    • GPT-6 SolonHealthBench Professional60.8

      OpenAI; length-adjusted; maximum reasoning effort; 60.8 (59.5, 1573) (adjusted, raw, mean response characters)

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench Professional length-adjusted | 51.8 (57.2, 3818) | 60.5 (64.1, 3228) | 57.7 (62.4, 3618) | 55.7 (59.8, 3389) | 64.7 (68.2, 3185) | 60.8 (59.5, 1573) | 60.8 (61.2, 2119)”

      Section 11.4.1, Table 29, HealthBench Professional length-adjusted, GPT-6 Sol column; September 22 correction.

    • GPT-6 AstraonHealthBench58.3

      length-adjusted, max reasoning effort (59.7 unadjusted, 2,258 mean response chars); GPT-6 Astra system card Table 6, column 'gpt-6 Astra'.

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench length-adjusted | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930) | 58.3 (56.9, 1760) | 53.2 (47.1, 977) | 54.5 (50.0, 1255)”

      Section 11.4.1, Table 29, HealthBench length-adjusted, GPT-6 Astra column; September 22 correction.

    • GPT-6 LunaonHealthBench54.5

      OpenAI; length-adjusted; maximum reasoning effort; 54.5 (50.0, 1255) (adjusted, raw, mean response characters)

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench length-adjusted | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930) | 58.3 (56.9, 1760) | 53.2 (47.1, 977) | 54.5 (50.0, 1255)”

      Section 11.4.1, Table 29, HealthBench length-adjusted, GPT-6 Luna column; September 22 correction.

    • GPT-6 SolonHealthBench53.2

      OpenAI; length-adjusted; maximum reasoning effort; 53.2 (47.1, 977) (adjusted, raw, mean response characters)

      “Evaluation | GPT-5.5 | GPT-5.6 Sol | GPT-5.6 Terra | GPT-5.6 Luna | GPT-6 Astra | GPT-6 Sol | GPT-6 Luna HealthBench length-adjusted | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930) | 58.3 (56.9, 1760) | 53.2 (47.1, 977) | 54.5 (50.0, 1255)”

      Section 11.4.1, Table 29, HealthBench length-adjusted, GPT-6 Sol column; September 22 correction.

  8. Claude Fable 5.1 and Claude Mythos 5.1 System Card System card from Anthropic, published 1 September 2026, checked 7 September 2026. First-party.

    8 results from this document
    • Claude Fable 5onHealthBench Professional63.3%

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 70.3%). Measured as Claude Mythos 5; Anthropic itself prints 66.0 in the Fable 5 column of the Opus 5 card with footnote Mythos 5.

      This document disagrees with the primary source.

      “HealthBench Professional 62.1% 63.3% 59.8% –”

      p. 167, Table 8.1.A, column 'Claude Fable 5/Mythos 5'

    • Claude Fable 5 (September card)onHealthBench Professional63.3%

      Measured as Claude Fable 5; Anthropic September card; length-adjusted; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt; raw 68.9%

      “HealthBench Professional | Length-adjusted score | Claude Fable 5 | 63.3%”

      p. 199, Figure 8.17.2.A, Claude Fable 5 length-adjusted bar (visually read printed labels).

    • Claude Fable 5.1onHealthBench Professional62.1%

      length-adjusted (method published in the HealthBench Professional paper); Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over five trials, no tools or customized system prompt; Fable 5.1 run with safety classifiers active and a refusal-fallback to Claude Opus 5 (raw 74.2%).

      “On HealthBench Professional, Claude Fable 5.1 achieved a raw score of 74.2%, ahead of Claude Opus 5 at 73.4%, Fable 5 at 68.9%, and Claude Sonnet 5 at 62.4%. After length adjustment, which penalizes verbose model responses, Fable 5.1 achieved a score of 62.1%.”

      p. 199, sec. 8.17.2 (same figure printed as 62.1% in Table 8.1.A, p. 167)

    • Claude Opus 5onHealthBench Professional59.8%

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 73.4%).

      This document repeats the primary source.

      “HealthBench Professional 62.1% 63.3% 59.8% –”

      p. 167, Table 8.1.A, column 'Claude Opus 5'; raw 73.4% also reprinted on p. 199, sec. 8.17.2

    • Claude Opus 5onHealthBench67.1%

      raw/unadjusted, Anthropic system-card protocol (max effort, no tools, five trials); length-adjusted 57.8; mirrored on benchlm.ai

      This document repeats the primary source.

      “On HealthBench, Claude Fable 5.1 achieved a raw score of 66.7%, ahead of Claude Fable 5 at 61.2% and Claude Sonnet 5 at 59.2%, and behind Claude Opus 5 at 67.1%.”

      p. 198, sec. 8.17.1

    • Claude Fable 5onHealthBench61.2% (raw)

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (40k token budget); printed in the Mythos 5 column of Table 8.1.A (Fable 5 column '-'); raw 62.5% per Opus 5 card

      This document disagrees with the primary source.

      “On HealthBench, Claude Fable 5.1 achieved a raw score of 66.7%, ahead of Claude Fable 5 at 61.2% and Claude Sonnet 5 at 59.2%, and behind Claude Opus 5 at 67.1%.”

      p. 198, sec. 8.17.1

    • Claude Fable 5onHealthBench60.4%

      Measured as Claude Fable 5; Anthropic September card; length-adjusted; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt; raw 61.2%

      “HealthBench | Length-adjusted score | Claude Fable 5 | 60.4%”

      p. 198, Figure 8.17.1.A, Claude Fable 5 length-adjusted bar (visually read printed labels).

    • Claude Fable 5.1onHealthBench60%

      length-adjusted (method published in OpenAI's GPT-5.5 System Card); Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over five trials, no tools or customized system prompt; Fable 5.1 run with safety classifiers active and a refusal-fallback to Claude Opus 5 (raw 66.7%).

      “On HealthBench, Claude Fable 5.1 achieved a raw score of 66.7%, ahead of Claude Fable 5 at 61.2% and Claude Sonnet 5 at 59.2%, and behind Claude Opus 5 at 67.1%. After length adjustment, which penalizes verbose model responses, Fable 5.1 achieved a score of 60%.”

      p. 198, sec. 8.17.1 / Figure 8.17.1.A

  9. GPT-5.6 - August Updates (system card addendum) System card from OpenAI, published 6 August 2026, checked 7 September 2026. First-party.

    6 results from this document
    • GPT-5.6 Sol (August)onHealthBench Professional54.0

      ChatGPT production (Instant) deployment setting, length-adjusted, GPT-5.6 August Updates PDF column GPT-5.6 Sol (August) (56.6 unadjusted, 2,894 chars)

      “HealthBench Professional 32.9 (33.8, 2,285) 38.4 (40.7, 2,775) 54.0 (56.6, 2,894) 44.1 (46.8, 2,920)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.6 Sol (August)

    • GPT-5.6 Luna (August)onHealthBench Professional44.1

      ChatGPT production (Instant) deployment setting, length-adjusted, GPT-5.6 August Updates PDF column GPT-5.6 Luna (August) (46.8 unadjusted, 2,920 chars)

      “HealthBench Professional 32.9 (33.8, 2,285) 38.4 (40.7, 2,775) 54.0 (56.6, 2,894) 44.1 (46.8, 2,920)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.6 Luna (August)

    • GPT-5.5 InstantonHealthBench Professional38.4

      length-adjusted (40.7 unadjusted, 2,775 mean response chars); GPT-5.5 Instant system card Table 5, column GPT-5.5 INSTANT.

      This document repeats the primary source.

      “HealthBench Professional 32.9 (33.8, 2,285) 38.4 (40.7, 2,775) 54.0 (56.6, 2,894) 44.1 (46.8, 2,920)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.5 Instant

    • GPT-5.6 Sol (August)onHealthBench55.0

      ChatGPT production/Instant deployment setting, length-adjusted (52.1 unadjusted), GPT-5.6 August Updates PDF 2026-08-06

      “HealthBench 49.6 (47.9, 1,724) 51.4 (50.9, 1,922) 55.0 (52.1, 1,514) 53.3 (50.7, 1,567)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.6 Sol (August)

    • GPT-5.6 Luna (August)onHealthBench53.3

      ChatGPT production (Instant) deployment setting, length-adjusted, GPT-5.6 August Updates PDF column GPT-5.6 Luna (August) (50.7 unadjusted, 1,567 chars)

      “HealthBench 49.6 (47.9, 1,724) 51.4 (50.9, 1,922) 55.0 (52.1, 1,514) 53.3 (50.7, 1,567)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.6 Luna (August)

    • GPT-5.5 InstantonHealthBench51.4

      length-adjusted, GPT-5.5 Instant system card Table 5 column GPT-5.5 INSTANT (50.9 unadjusted, 1,922 chars); same number in GPT-5.6 August Updates p. 11

      This document repeats the primary source.

      “HealthBench 49.6 (47.9, 1,724) 51.4 (50.9, 1,922) 55.0 (52.1, 1,514) 53.3 (50.7, 1,567)”

      p. 11, section 5.1 HealthBench, table "Reported as length-adjusted score (unadjusted, mean response length in characters)", column GPT-5.5 Instant

  10. System Card: Claude Opus 5 System card from Anthropic, published 24 July 2026, checked 7 September 2026. First-party.

    9 results from this document
    • Claude Fable 5onHealthBench Professional66.0

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 70.3%). Measured as Claude Mythos 5; Anthropic itself prints 66.0 in the Fable 5 column of the Opus 5 card with footnote Mythos 5.

      This document repeats the primary source.

      “HealthBench Professional 59.8 57.4 66.011 60.5”

      p. 152, Table 8.1.A, column Fable 5 (footnote 11: Mythos 5)

    • Claude Opus 5onHealthBench Professional59.8%

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 73.4%).

      “Claude Opus 5 achieved a raw score of 73.4%, which is the highest amongst all Claude models, ahead of Claude Mythos 5 at 70.3%, Claude Opus 4.8 at 60.3%, and Claude Sonnet 5 at 62.4%. After length adjustment, which penalizes verbose model responses, Claude Opus 5 achieved a score of 59.8%.”

      p. 189, section 8.15.2 HealthBench Professional results; also Table 8.1.A p. 152 ('HealthBench Professional 59.8 ...')

    • Claude Sonnet 5onHealthBench Professional57.8%

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 62.4%).

      This document repeats the primary source.

      “Figure 8.15.2.A bar labels: Claude Sonnet 5 62.4% raw / 57.8% length-adjusted (prose: 'Claude Sonnet 5 at 62.4%')”

      p. 189, section 8.15.2, Figure 8.15.2.A

    • Claude Opus 4.8onHealthBench Professional57.4

      length-adjusted, adaptive thinking at max effort, Claude Sonnet 4.6 grader (the grader used by the Opus 4.8 card itself); the other Claude rows on this board use the Claude Opus 4.8 grader, under which Anthropic later prints 57.4 for this model.

      This document disagrees with the primary source.

      “HealthBench Professional 59.8 57.4 66.011 60.5”

      p. 152, Table 8.1.A, column Opus 4.8

    • Claude Opus 5onHealthBench67.1

      raw/unadjusted, Anthropic system-card protocol (max effort, no tools, five trials); length-adjusted 57.8; mirrored on benchlm.ai

      “Claude Opus 5 achieved a raw score of 67.1%, which is the highest among all Claude models, ahead of Claude Mythos 5 at 62.5%, Claude Opus 4.8 at 58.8%, and Claude Sonnet 5 at 59.2%. After length adjustment, which penalizes verbose model responses, Claude Opus 5 achieved a score of 57.8%.”

      p. 188, section 8.15.1 HealthBench results; Figure 8.15.1.A

    • Claude Fable 5onHealthBench62.7%

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (40k token budget); printed in the Mythos 5 column of Table 8.1.A (Fable 5 column '-'); raw 62.5% per Opus 5 card

      This document repeats the primary source.

      “Figure 8.15.1.A bar labels: Claude Mythos 5 62.5% raw / 62.7% length-adjusted (prose: 'Claude Mythos 5 at 62.5%')”

      p. 188, section 8.15.1, Figure 8.15.1.A

    • Claude Opus 4.8onHealthBench59.3%

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; comparison column in the Fable/Mythos 5 card; raw 58.8% per Opus 5 card; not in the Opus 4.8 card itself

      This document repeats the primary source.

      “Figure 8.15.1.A bar labels: Claude Opus 4.8 58.8% raw / 59.3% length-adjusted (prose: 'Claude Opus 4.8 at 58.8%')”

      p. 188, section 8.15.1, Figure 8.15.1.A

    • Claude Sonnet 5onHealthBench58.7%

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; figure-only in the Sonnet 5 card; raw 59.2% per Opus 5 card

      This document repeats the primary source.

      “Figure 8.15.1.A bar labels: Claude Sonnet 5 59.2% raw / 58.7% length-adjusted (prose: 'Claude Sonnet 5 at 59.2%')”

      p. 188, section 8.15.1, Figure 8.15.1.A

    • Claude Opus 5 (length-adjusted)onHealthBench57.8%

      Anthropic; length-adjusted; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt; raw 67.1%

      “8.15.1 HealthBench results After length adjustment, which penalizes verbose model responses, Claude Opus 5 achieved a score of 57.8%.”

      p. 188, Section 8.15.1, paragraph and Figure 8.15.1.A.

  11. GPT-5.6 System Card System card from OpenAI, published 9 July 2026, checked 7 September 2026. First-party.

    17 results from this document
    • GPT-5.6 SolonHealthBench Professional60.5

      length-adjusted, max reasoning effort (64.1 unadjusted, 3,228 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-SOL. API max-effort setting, not the August ChatGPT production setting measured in row 492.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-SOL

    • GPT-5.6 TerraonHealthBench Professional57.7

      length-adjusted, max reasoning effort (62.4 unadjusted, 3,618 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-TERRA.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-TERRA

    • GPT-5.6 LunaonHealthBench Professional55.7

      length-adjusted, max reasoning effort (59.8 unadjusted, 3,389 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-LUNA. API max-effort setting, not the August ChatGPT production setting measured in row 493.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-LUNA

    • GPT-5.5onHealthBench Professional51.8

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.5 (57.2 unadjusted, 3818 chars)

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.5

    • GPT-5.4onHealthBench Professional48.1

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.4 (51.9 unadjusted, 3308 chars)

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.4

    • GPT-5onHealthBench Professional46.2

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5 (51.0 unadjusted, 3616 chars)

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5

    • GPT-5.2onHealthBench Professional45.9

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.2 (50.0 unadjusted, 3400 chars)

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.2

    • GPT-5.1onHealthBench Professional39.6

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.1 (48.0 unadjusted, 4863 chars)

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.1

    • GPT-5.2-HighonHealthBench56.8

      raw score as run by Baichuan in the M3 technical report, not an OpenAI-reported number; OpenAI own GPT-5.2 figure is 56.8 length-adjusted (60.7 unadjusted) in the GPT-5.6 system card.

      This document disagrees with the primary source.

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.2

    • GPT-5onHealthBench57.7

      OpenAI; length-adjusted; maximum reasoning effort; 57.7 (63.1, 2904) (adjusted, raw, mean response characters)

      “evaluation | gpt-5 | gpt-5.1 | gpt-5.2 | gpt-5.4 | gpt-5.5 | gpt-5.6-sol | gpt-5.6-terra | gpt-5.6-luna HealthBench length-adjusted | 57.7 (63.1, 2904) | 50.9 (64.2, 4222) | 56.8 (60.7, 2645) | 54.0 (55.7, 2275) | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930)”

      Section 6.1, Table 6, HealthBench length-adjusted row, gpt-5 column.

    • GPT-5.6 SolonHealthBench57.0

      length-adjusted, max reasoning effort (55.6 unadjusted), GPT-5.6 system card 2026-07-09

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-SOL

    • GPT-5.6 TerraonHealthBench57.0

      length-adjusted (58.7 unadjusted), max reasoning effort

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-TERRA

    • GPT-5.2onHealthBench56.8

      OpenAI; length-adjusted; maximum reasoning effort; 56.8 (60.7, 2645) (adjusted, raw, mean response characters)

      “evaluation | gpt-5 | gpt-5.1 | gpt-5.2 | gpt-5.4 | gpt-5.5 | gpt-5.6-sol | gpt-5.6-terra | gpt-5.6-luna HealthBench length-adjusted | 57.7 (63.1, 2904) | 50.9 (64.2, 4222) | 56.8 (60.7, 2645) | 54.0 (55.7, 2275) | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930)”

      Section 6.1, Table 6, HealthBench length-adjusted row, gpt-5.2 column.

    • GPT-5.5onHealthBench56.5

      length-adjusted (58.4 unadjusted), comparison row in GPT-5.6 system card

      This document repeats the primary source.

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.5

    • GPT-5.6 LunaonHealthBench55.8

      length-adjusted (55.4 unadjusted), max reasoning effort

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.6-LUNA

    • GPT-5.4onHealthBench54.0

      OpenAI; length-adjusted; maximum reasoning effort; 54.0 (55.7, 2275) (adjusted, raw, mean response characters)

      “evaluation | gpt-5 | gpt-5.1 | gpt-5.2 | gpt-5.4 | gpt-5.5 | gpt-5.6-sol | gpt-5.6-terra | gpt-5.6-luna HealthBench length-adjusted | 57.7 (63.1, 2904) | 50.9 (64.2, 4222) | 56.8 (60.7, 2645) | 54.0 (55.7, 2275) | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930)”

      Section 6.1, Table 6, HealthBench length-adjusted row, gpt-5.4 column.

    • GPT-5.1onHealthBench50.9

      OpenAI; length-adjusted; maximum reasoning effort; 50.9 (64.2, 4222) (adjusted, raw, mean response characters)

      “evaluation | gpt-5 | gpt-5.1 | gpt-5.2 | gpt-5.4 | gpt-5.5 | gpt-5.6-sol | gpt-5.6-terra | gpt-5.6-luna HealthBench length-adjusted | 57.7 (63.1, 2904) | 50.9 (64.2, 4222) | 56.8 (60.7, 2645) | 54.0 (55.7, 2275) | 56.5 (58.4, 2313) | 57.0 (55.6, 1764) | 57.0 (58.7, 2285) | 55.8 (55.4, 1930)”

      Section 6.1, Table 6, HealthBench length-adjusted row, gpt-5.1 column.

  12. System Card: Claude Sonnet 5 System card from Anthropic, published 30 June 2026, checked 7 September 2026. First-party.

    4 results from this document
    • Claude Sonnet 5onHealthBench Professional57.8

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 62.4%).

      “HealthBench Professional 57.8 44.2 51.8 -”

      p. 115, Table 8.1.A, row HealthBench Professional, column Claude Sonnet 5; Figure 8.12.2.A p. 139

    • Claude Opus 4.8 (Opus 4.8 grader)onHealthBench Professional57.4%

      Anthropic June evaluation; length-adjusted; adaptive max effort; Opus 4.8 grader; five trials; no tools or custom system prompt

      “HealthBench Professional | Length-adjusted score (%) | Claude Opus 4.8 | 57.4%”

      p. 139, Figure 8.12.2.A, Claude Opus 4.8 bar (visually read printed labels).

    • Claude Sonnet 4.6onHealthBench Professional44.2

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; comparison column in the Sonnet 5 card. Other Anthropic prints: 44.4% (Fable card Figure 8.18.2.A), 41.7% (Opus 4.8 card, Sonnet 4.6 grader)

      “HealthBench Professional 57.8 44.2 51.8 -”

      p. 115, Table 8.1.A, row HealthBench Professional, column Claude Sonnet 4.6

    • Claude Sonnet 5onHealthBench58.7%

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; figure-only in the Sonnet 5 card; raw 59.2% per Opus 5 card

      “[Figure 8.12.1.A] HealthBench length-adjusted scores. All Claude models used adaptive thinking at max effort. (bar label: Claude Sonnet 5 58.7%)”

      p. 138, section 8.12.1 HealthBench results, Figure 8.12.1.A bar label (no prose or table number)

  13. GPT-5.6 Preview System Card System card from OpenAI, published 26 June 2026, checked 7 September 2026. First-party.

    14 results from this document
    • GPT-5.6 SolonHealthBench Professional60.5

      length-adjusted, max reasoning effort (64.1 unadjusted, 3,228 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-SOL. API max-effort setting, not the August ChatGPT production setting measured in row 492.

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-SOL

    • GPT-5.6 TerraonHealthBench Professional57.7

      length-adjusted, max reasoning effort (62.4 unadjusted, 3,618 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-TERRA.

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-TERRA

    • GPT-5.6 LunaonHealthBench Professional55.7

      length-adjusted, max reasoning effort (59.8 unadjusted, 3,389 mean response chars); GPT-5.6 system card Table 6, column GPT-5.6-LUNA. API max-effort setting, not the August ChatGPT production setting measured in row 493.

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-LUNA

    • GPT-5.5onHealthBench Professional51.8

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.5 (57.2 unadjusted, 3818 chars)

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.5

    • GPT-5.4onHealthBench Professional48.1

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.4 (51.9 unadjusted, 3308 chars)

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.4

    • GPT-5onHealthBench Professional46.2

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5 (51.0 unadjusted, 3616 chars)

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5

    • GPT-5.2onHealthBench Professional45.9

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.2 (50.0 unadjusted, 3400 chars)

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.2

    • GPT-5.1onHealthBench Professional39.6

      length-adjusted, max reasoning effort, GPT-5.6 system card Table 6 column GPT-5.1 (48.0 unadjusted, 4863 chars)

      This document repeats the primary source.

      “HealthBench Professional length-adjusted 46.2 (51.0, 3616) 39.6 (48.0, 4863) 45.9 (50.0, 3400) 48.1 (51.9, 3308) 51.8 (57.2, 3818) 60.5 (64.1, 3228) 57.7 (62.4, 3618) 55.7 (59.8, 3389)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.1

    • Claude Opus 5onHealthBench67.1

      raw/unadjusted, Anthropic system-card protocol (max effort, no tools, five trials); length-adjusted 57.8; mirrored on benchlm.ai

      This document repeats the primary source.

    • GPT-5.6 SolonHealthBench57.0

      length-adjusted, max reasoning effort (55.6 unadjusted), GPT-5.6 system card 2026-07-09

      This document repeats the primary source.

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-SOL

    • GPT-5.6 TerraonHealthBench57.0

      length-adjusted (58.7 unadjusted), max reasoning effort

      This document repeats the primary source.

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-TERRA

    • GPT-5.5onHealthBench56.5

      length-adjusted (58.4 unadjusted), comparison row in GPT-5.6 system card

      This document repeats the primary source.

    • GPT-5.6 LunaonHealthBench55.8

      length-adjusted (55.4 unadjusted), max reasoning effort

      This document repeats the primary source.

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313) 57.0 (55.6, 1764) 57.0 (58.7, 2285) 55.8 (55.4, 1930)”

      Section 5.1 HealthBench, Table 6 (reported as length-adjusted score (unadjusted, mean response length in characters)) (preview card, published 2026-06-26), column GPT-5.6-LUNA

    • GPT-5.6 Sol (August)onHealthBench55.0

      ChatGPT production/Instant deployment setting, length-adjusted (52.1 unadjusted), GPT-5.6 August Updates PDF 2026-08-06

      This document repeats the primary source.

  14. Claude Fable 5 and Claude Mythos 5 System Card System card from Anthropic, published 9 June 2026, checked 7 September 2026. First-party.

    4 results from this document
    • Claude Fable 5onHealthBench Professional66.0

      length-adjusted, Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (raw 70.3%). Measured as Claude Mythos 5; Anthropic itself prints 66.0 in the Fable 5 column of the Opus 5 card with footnote Mythos 5.

      “HealthBench Professional 66.0 - 64.7 56.9 51.8 -”

      p. 252, Table 8.1.A, row HealthBench Professional, column Mythos 5 (Fable 5 column '-'); Figure 8.18.2.A p. 298 bar label 66.0% on Claude Mythos 5

    • Claude Opus 4.8onHealthBench Professional56.9

      length-adjusted, adaptive thinking at max effort, Claude Sonnet 4.6 grader (the grader used by the Opus 4.8 card itself); the other Claude rows on this board use the Claude Opus 4.8 grader, under which Anthropic later prints 57.4 for this model.

      This document disagrees with the primary source.

      “HealthBench Professional 66.0 - 64.7 56.9 51.8 -”

      p. 252, Table 8.1.A, column Opus 4.8

    • Claude Fable 5onHealthBench62.7

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt (40k token budget); printed in the Mythos 5 column of Table 8.1.A (Fable 5 column '-'); raw 62.5% per Opus 5 card

      “HealthBench 62.7 - 61.1 59.3 56.5 -”

      p. 252, Table 8.1.A, row HealthBench, column Mythos 5; Figure 8.18.1.A p. 297

    • Claude Opus 4.8onHealthBench59.3

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; comparison column in the Fable/Mythos 5 card; raw 58.8% per Opus 5 card; not in the Opus 4.8 card itself

      “HealthBench 62.7 - 61.1 59.3 56.5 -”

      p. 252, Table 8.1.A, row HealthBench, column Opus 4.8

  15. System Card: Claude Opus 4.8 System card from Anthropic, published 28 May 2026, checked 7 September 2026. First-party.

    3 results from this document
    • Claude Opus 4.8onHealthBench Professional55.8%

      length-adjusted, adaptive thinking at max effort, Claude Sonnet 4.6 grader (the grader used by the Opus 4.8 card itself); the other Claude rows on this board use the Claude Opus 4.8 grader, under which Anthropic later prints 57.4 for this model.

      “Claude Opus 4.8 scores 55.8%, a meaningful improvement over Claude Opus 4.7 at 51.9% and Claude Sonnet 4.6 at 41.7%.”

      p. 228, section 8.14.1 HealthBench Professional; Figure 8.14.A p. 229

    • Claude Opus 4.7onHealthBench Professional51.9%

      length-adjusted; adaptive thinking at max effort; Claude Sonnet 4.6 grader; 5 trials; comparison model in the Opus 4.8 card

      “Claude Opus 4.8 scores 55.8%, a meaningful improvement over Claude Opus 4.7 at 51.9% and Claude Sonnet 4.6 at 41.7%.”

      p. 228, section 8.14.1 HealthBench Professional; Figure 8.14.A p. 229

    • Claude Sonnet 4.6onHealthBench Professional41.7%

      length-adjusted; Anthropic protocol: adaptive thinking at max effort, Claude Opus 4.8 grader, averaged over 5 trials, no tools or custom system prompt; comparison column in the Sonnet 5 card. Other Anthropic prints: 44.4% (Fable card Figure 8.18.2.A), 41.7% (Opus 4.8 card, Sonnet 4.6 grader)

      This document disagrees with the primary source.

      “Claude Opus 4.8 scores 55.8%, a meaningful improvement over Claude Opus 4.7 at 51.9% and Claude Sonnet 4.6 at 41.7%.”

      p. 228, section 8.14.1

  16. GPT-5.5 Instant System Card System card from OpenAI, published 5 May 2026, checked 7 September 2026. First-party.

    3 results from this document
    • GPT-5.5 InstantonHealthBench Professional38.4

      length-adjusted (40.7 unadjusted, 2,775 mean response chars); GPT-5.5 Instant system card Table 5, column GPT-5.5 INSTANT.

      “HealthBench Professional 37.6 (40.4, 2,973) 35.7 (38.3, 2,872) 32.9 (33.8, 2,285) 38.4 (40.7, 2,775)”

      Section 4.1 HealthBench, Table 5 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.5 INSTANT

    • GPT-5.3 ChatonHealthBench49.6

      raw score (no length adjustment), GPT-5.3 Instant system card Table 3 column GPT-5.3-INSTANT; later OpenAI cards print 49.6 length-adjusted (47.9 unadjusted)

      This document disagrees with the primary source.

      “HealthBench 49.6 (50.8, 2,208) 50.6 (51.5, 2,145) 49.6 (47.9, 1,724) 51.4 (50.9, 1,922)”

      Section 4.1 HealthBench, Table 5 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.3 INSTANT

    • GPT-5.5 InstantonHealthBench51.4

      length-adjusted, GPT-5.5 Instant system card Table 5 column GPT-5.5 INSTANT (50.9 unadjusted, 1,922 chars); same number in GPT-5.6 August Updates p. 11

      “HealthBench 49.6 (50.8, 2,208) 50.6 (51.5, 2,145) 49.6 (47.9, 1,724) 51.4 (50.9, 1,922)”

      Section 4.1 HealthBench, Table 5 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.5 INSTANT

  17. GPT-5.5 System Card System card from OpenAI, published 23 April 2026, checked 7 September 2026. First-party.

    1 result from this document
    • GPT-5.5onHealthBench56.5

      length-adjusted (58.4 unadjusted), comparison row in GPT-5.6 system card

      “HealthBench length-adjusted 57.7 (63.1, 2904) 50.9 (64.2, 4222) 56.8 (60.7, 2645) 54.0 (55.7, 2275) 56.5 (58.4, 2313)”

      Section 5 Health, Table 7 (reported as length-adjusted score (unadjusted, mean response length in characters)), column GPT-5.5

  18. GPT-5.3 Instant System Card System card from OpenAI, published 2 March 2026, checked 7 September 2026. First-party.

    1 result from this document
    • GPT-5.3 ChatonHealthBench54.1%

      raw score (no length adjustment), GPT-5.3 Instant system card Table 3 column GPT-5.3-INSTANT; later OpenAI cards print 49.6 length-adjusted (47.9 unadjusted)

      “HealthBench 55.4% 54.1%”

      Section 4.1 HealthBench, Table 3: HealthBench, row HealthBench, column GPT-5.3-INSTANT

Model cards

  1. Muse Wonder Model Card Model card from Meta, published 12 October 2026, checked 2 November 2026. First-party.

    2 results from this document
    • Muse Wonder FullonHealthBench Professional62.3

      Meta internal run; length-normalized weighted rubric score, GPT-5.4 grader, no tools, max reasoning effort (Muse Wonder Model Card p. 3 table, Muse Wonder Full (max) column)

      “HealthBench Pro (Open-ended health queries) | Muse Wonder Full (max) 62.3 | Muse Wonder (max) 60.5 | Claude Fable 5.1 (max) 62.1 | GPT-6 Astra (max) 65.3 | Claude Opus 5.5 (max) 65.6 Methodology (p. 4): HealthBench Pro evaluates responses to 525 professional-grade health questions. Models answer without tools, and a GPT-5.4 grader scores each response against physician-written rubrics. We run the official tasks and grading in a common internal framework. The primary metric is the length-normalized weighted rubric score.”

      p. 3, results table (image), HealthBench Pro row, Muse Wonder Full (max) column; methodology p. 4

    • Muse WonderonHealthBench Professional60.5

      Meta internal run; length-normalized weighted rubric score, GPT-5.4 grader, no tools, max reasoning effort (Muse Wonder Model Card p. 3 table, Muse Wonder (max) column)

      “HealthBench Pro (Open-ended health queries) | Muse Wonder Full (max) 62.3 | Muse Wonder (max) 60.5 | Claude Fable 5.1 (max) 62.1 | GPT-6 Astra (max) 65.3 | Claude Opus 5.5 (max) 65.6 Methodology (p. 4): HealthBench Pro evaluates responses to 525 professional-grade health questions. Models answer without tools, and a GPT-5.4 grader scores each response against physician-written rubrics. We run the official tasks and grading in a common internal framework. The primary metric is the length-normalized weighted rubric score.”

      p. 3, results table (image), HealthBench Pro row, Muse Wonder (max) column; methodology p. 4

  2. MAI-Thinking-1: Building a Hill-Climbing Machine Model card from Microsoft AI, published 12 August 2026, checked 7 September 2026. First-party.

    1 result from this document
    • MAI-Thinking-1onHealthBench Professional35

      length-adjusted (HealthBench Professional length penalty), standard GPT-5.4 grader and OpenAI rubrics, Microsoft AI run; printed at integer precision.

      “Model AIR-Bench CyberSec Instruct CyberSec Auto Long Fact Truthful QA HealthBench Prof. MedXpert QA MAI-Thinking-1 88 63 63 98 88 35 43 Sonnet 4.6 88 62 56 98 88 38 49”

      p. 54, Table 12 'Post-trained model evaluation results on various public benchmarks', Health group, column HealthBench Prof.; protocol Appendix K.6 p. 106: HealthBench Professional introduces a length penalty for the primary metric, to correct for a well-observed correlation between lengthy responses and artificially increased LLM-grader scores. For all reported scores, we use the standard GPT-5.4 grader and rubrics provided by OpenAI.

  3. Muse Spark 1.1 Evaluation Report Model card from Meta, published 9 July 2026, checked 7 September 2026. First-party.

    2 results from this document
    • Muse Spark 1.1onHealthBench Professional59.3

      length-normalized, GPT-5.4 low-reasoning grader, xhigh reasoning via Meta Model API (Muse Spark 1.1 Evaluation Report Figure 44)

      “Health | HealthBench Professional | 59.3 | 54.1 | 41.6 | 55.8 | 51.8 (Figure 44 image table row; columns Muse Spark 1.1, Muse Spark, Gemini 3.1 Pro (high), Opus 4.8 (max), GPT 5.5 (xhigh))”

      p. 101, Figure 44 'General capability benchmark results' (image), row HealthBench Professional, column Muse Spark 1.1; protocol p. 104 (printed 103): HealthBench Pro comprises 525 evaluation data points graded by rubrics. We use GPT-5.4 with low reasoning effort as the grader and report the length-normalized rubric score as done in their paper.

    • Muse SparkonHealthBench Professional54.1

      length-normalized, GPT-5.4 low-reasoning grader; Muse Spark (1.0) column in the Muse Spark 1.1 Evaluation Report Figure 44

      “Health | HealthBench Professional | 59.3 | 54.1 | 41.6 | 55.8 | 51.8 (Figure 44 image table row; columns Muse Spark 1.1, Muse Spark, Gemini 3.1 Pro (high), Opus 4.8 (max), GPT 5.5 (xhigh))”

      p. 101, Figure 44 (image), row HealthBench Professional, column Muse Spark; protocol p. 104 (printed 103)

  4. gpt-oss-120b & gpt-oss-20b Model Card Model card from OpenAI, published 5 August 2025, checked 7 September 2026. First-party.

    2 results from this document
    • GPT OSS 120BonHealthBench57.6

      reasoning level high, raw score (%), gpt-oss model card Table 3 (low 53.0, medium 55.9)

      “HealthBench 53.0 55.9 57.6 40.4 41.8 42.5”

      Section 2, Table 3: Evaluations across multiple benchmarks and reasoning levels, row HealthBench, column gpt-oss-120b high

    • GPT OSS 20BonHealthBench42.5

      reasoning level high, raw score (%), gpt-oss model card Table 3 (low 40.4, medium 41.8)

      “HealthBench 53.0 55.9 57.6 40.4 41.8 42.5”

      Section 2, Table 3: Evaluations across multiple benchmarks and reasoning levels, row HealthBench, column gpt-oss-20b high

Official leaderboards

  1. SAFE-Teen, Vals AI Official leaderboard from Vals AI, published 13 October 2026. First-party.

    54 results from this document
    • GPT-6.1 SolonSAFE-Teen (Vals AI)15%

      no teen-aware prompt

      “GPT-6.1 Sol | 15% | 9% | 19% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Sonnet 5.5onSAFE-Teen (Vals AI)23%

      no teen-aware prompt

      “Claude Sonnet 5.5 | 23% | 43% | 37% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GPT-5.6 LunaonSAFE-Teen (Vals AI)23%

      no teen-aware prompt

      “GPT-5.6 Luna | 23% | 21% | 35% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Extra High)onSAFE-Teen (Vals AI)26%

      no teen-aware prompt, Extra High

      “Muse Spark 1.3 (Extra High) | 26% | 14% | 56% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Qwen 3.8 MaxonSAFE-Teen (Vals AI)26%

      no teen-aware prompt

      “Qwen 3.8 Max | 26% | 35% | 51% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Opus 5.5onSAFE-Teen (Vals AI)27%

      no teen-aware prompt

      “Claude Opus 5.5 | 27% | 44% | 25% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • DeepSeek V4.1 FlashonSAFE-Teen (Vals AI)31%

      no teen-aware prompt

      “DeepSeek V4.1 Flash | 31% | 41% | 51% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Mistral Large 4 PreviewonSAFE-Teen (Vals AI)33%

      no teen-aware prompt

      “Mistral Large 4 Preview | 33% | 53% | 19% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Minimal)onSAFE-Teen (Vals AI)33%

      no teen-aware prompt, Minimal

      “Muse Spark 1.3 (Minimal) | 33% | 29% | 70% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiMo v2.6 ProonSAFE-Teen (Vals AI)40%

      no teen-aware prompt

      “MiMo v2.6 Pro | 40% | 57% | 56% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.7onSAFE-Teen (Vals AI)44%

      no teen-aware prompt

      “Grok 4.7 | 44% | 39% | 48% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Kimi K3onSAFE-Teen (Vals AI)44%

      no teen-aware prompt

      “Kimi K3 | 44% | 61% | 63% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.5onSAFE-Teen (Vals AI)50%

      no teen-aware prompt

      “Grok 4.5 | 50% | 47% | 65% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GLM 5.3onSAFE-Teen (Vals AI)51%

      no teen-aware prompt

      “GLM 5.3 | 51% | 70% | 64% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.6 FlashonSAFE-Teen (Vals AI)52%

      no teen-aware prompt

      “Gemini 3.6 Flash | 52% | 68% | 70% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 4 ArgononSAFE-Teen (Vals AI)53%

      no teen-aware prompt

      “Gemini 4 Argon | 53% | 71% | 66% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiniMax M3onSAFE-Teen (Vals AI)54%

      no teen-aware prompt

      “MiniMax M3 | 54% | 49% | 53% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.8 FlashonSAFE-Teen (Vals AI)56%

      no teen-aware prompt

      “Gemini 3.8 Flash | 56% | 69% | 63% (Urgent safety | Health advice | Relationship boundaries)”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GPT-6.1 SolonSAFE-Teen (Vals AI)9%

      no teen-aware prompt

      “GPT-6.1 Sol | 15% | 9% | 19%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Extra High)onSAFE-Teen (Vals AI)14%

      no teen-aware prompt, Extra High

      “Muse Spark 1.3 (Extra High) | 26% | 14% | 56%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GPT-5.6 LunaonSAFE-Teen (Vals AI)21%

      no teen-aware prompt

      “GPT-5.6 Luna | 23% | 21% | 35%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Minimal)onSAFE-Teen (Vals AI)29%

      no teen-aware prompt, Minimal

      “Muse Spark 1.3 (Minimal) | 33% | 29% | 70%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Qwen 3.8 MaxonSAFE-Teen (Vals AI)35%

      no teen-aware prompt

      “Qwen 3.8 Max | 26% | 35% | 51%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.7onSAFE-Teen (Vals AI)39%

      no teen-aware prompt

      “Grok 4.7 | 44% | 39% | 48%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • DeepSeek V4.1 FlashonSAFE-Teen (Vals AI)41%

      no teen-aware prompt

      “DeepSeek V4.1 Flash | 31% | 41% | 51%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Sonnet 5.5onSAFE-Teen (Vals AI)43%

      no teen-aware prompt

      “Claude Sonnet 5.5 | 23% | 43% | 37%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Opus 5.5onSAFE-Teen (Vals AI)44%

      no teen-aware prompt

      “Claude Opus 5.5 | 27% | 44% | 25%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.5onSAFE-Teen (Vals AI)47%

      no teen-aware prompt

      “Grok 4.5 | 50% | 47% | 65%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiniMax M3onSAFE-Teen (Vals AI)49%

      no teen-aware prompt

      “MiniMax M3 | 54% | 49% | 53%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Mistral Large 4 PreviewonSAFE-Teen (Vals AI)53%

      no teen-aware prompt

      “Mistral Large 4 Preview | 33% | 53% | 19%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiMo v2.6 ProonSAFE-Teen (Vals AI)57%

      no teen-aware prompt

      “MiMo v2.6 Pro | 40% | 57% | 56%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Kimi K3onSAFE-Teen (Vals AI)61%

      no teen-aware prompt

      “Kimi K3 | 44% | 61% | 63%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.6 FlashonSAFE-Teen (Vals AI)68%

      no teen-aware prompt

      “Gemini 3.6 Flash | 52% | 68% | 70%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.8 FlashonSAFE-Teen (Vals AI)69%

      no teen-aware prompt

      “Gemini 3.8 Flash | 56% | 69% | 63%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GLM 5.3onSAFE-Teen (Vals AI)70%

      no teen-aware prompt

      “GLM 5.3 | 51% | 70% | 64%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 4 ArgononSAFE-Teen (Vals AI)71%

      no teen-aware prompt

      “Gemini 4 Argon | 53% | 71% | 66%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GPT-6.1 SolonSAFE-Teen (Vals AI)19%

      no teen-aware prompt

      “GPT-6.1 Sol | 15% | 9% | 19%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Mistral Large 4 PreviewonSAFE-Teen (Vals AI)19%

      no teen-aware prompt

      “Mistral Large 4 Preview | 33% | 53% | 19%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Opus 5.5onSAFE-Teen (Vals AI)25%

      no teen-aware prompt

      “Claude Opus 5.5 | 27% | 44% | 25%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GPT-5.6 LunaonSAFE-Teen (Vals AI)35%

      no teen-aware prompt

      “GPT-5.6 Luna | 23% | 21% | 35%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Claude Sonnet 5.5onSAFE-Teen (Vals AI)37%

      no teen-aware prompt

      “Claude Sonnet 5.5 | 23% | 43% | 37%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.7onSAFE-Teen (Vals AI)48%

      no teen-aware prompt

      “Grok 4.7 | 44% | 39% | 48%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • DeepSeek V4.1 FlashonSAFE-Teen (Vals AI)51%

      no teen-aware prompt

      “DeepSeek V4.1 Flash | 31% | 41% | 51%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Qwen 3.8 MaxonSAFE-Teen (Vals AI)51%

      no teen-aware prompt

      “Qwen 3.8 Max | 26% | 35% | 51%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiniMax M3onSAFE-Teen (Vals AI)53%

      no teen-aware prompt

      “MiniMax M3 | 54% | 49% | 53%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • MiMo v2.6 ProonSAFE-Teen (Vals AI)56%

      no teen-aware prompt

      “MiMo v2.6 Pro | 40% | 57% | 56%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Extra High)onSAFE-Teen (Vals AI)56%

      no teen-aware prompt, Extra High

      “Muse Spark 1.3 (Extra High) | 26% | 14% | 56%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.8 FlashonSAFE-Teen (Vals AI)63%

      no teen-aware prompt

      “Gemini 3.8 Flash | 56% | 69% | 63%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Kimi K3onSAFE-Teen (Vals AI)63%

      no teen-aware prompt

      “Kimi K3 | 44% | 61% | 63%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • GLM 5.3onSAFE-Teen (Vals AI)64%

      no teen-aware prompt

      “GLM 5.3 | 51% | 70% | 64%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Grok 4.5onSAFE-Teen (Vals AI)65%

      no teen-aware prompt

      “Grok 4.5 | 50% | 47% | 65%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 4 ArgononSAFE-Teen (Vals AI)66%

      no teen-aware prompt

      “Gemini 4 Argon | 53% | 71% | 66%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Gemini 3.6 FlashonSAFE-Teen (Vals AI)70%

      no teen-aware prompt

      “Gemini 3.6 Flash | 52% | 68% | 70%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

    • Muse Spark 1.3 (Minimal)onSAFE-Teen (Vals AI)70%

      no teen-aware prompt, Minimal

      “Muse Spark 1.3 (Minimal) | 33% | 29% | 70%”

      vals.ai/benchmarks/mental-health, results table (No teen-aware prompt)

  2. Best AI for Healthcare & Medical: LLM Leaderboard Official leaderboard from Artificial Analysis, checked 8 September 2026. First-party.

    29 results from this document
    • Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)onArtificial Analysis Healthcare & Medical Index61

      “Claude Opus 5.5 (max with fallback) | 61”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Claude Opus 5.5 (max with fallback)"; 25 input configurations explicitly selected (25 of 80 models); models slug=claude-opus-5-5; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)onArtificial Analysis Healthcare & Medical Index58

      “Claude Fable 5.1 (max with fallback) | 58”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Claude Fable 5.1 (max with fallback)"; 25 input configurations explicitly selected (25 of 80 models); models slug=claude-fable-5-1; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Claude Opus 5 (Adaptive Reasoning, Max Effort)onArtificial Analysis Healthcare & Medical Index53

      all underlying benchmarks run independently by Artificial Analysis

      “Claude Opus 5 (max) | 53”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Claude Opus 5 (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=claude-opus-5; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)onArtificial Analysis Healthcare & Medical Index52

      “Based on the Artificial Analysis Healthcare & Medical Index, the top-performing AI models for healthcare and medical work are currently Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) (56), Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) (52), and Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) (52). Rankings are updated as new models are released.”

      embedded JSON-LD FAQPage, answer to "Which AI is best for doctors and clinicians?"; the Score chart JSON-LD carries the same entry as {"label":"Claude Fable 5 (with fallback)","score":52.01006901640588,"detailsUrl":"/models/claude-fable-5"}; chart header "23 of 72 models"; retrieved 2026-09-07

    • GPT-6 Astra (max)onArtificial Analysis Healthcare & Medical Index52

      “GPT-6 Astra (max) | 52”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GPT-6 Astra (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gpt-6-astra; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Muse Spark 1.3 (max)onArtificial Analysis Healthcare & Medical Index50

      “Muse Spark 1.3 (max) | 50”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Muse Spark 1.3 (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=muse-spark-1-3; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • GLM-5.3 (max)onArtificial Analysis Healthcare & Medical Index47

      “GLM-5.3 (max) | 47”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GLM-5.3 (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=glm-5-3; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Grok 4.7 (xhigh)onArtificial Analysis Healthcare & Medical Index47

      “Grok 4.7 (xhigh) | 47”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Grok 4.7 (xhigh)"; 25 input configurations explicitly selected (25 of 80 models); models slug=grok-4-7; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • GLM 5.3 FlashonArtificial Analysis Healthcare & Medical Index45

      “GLM-5.3-Flash | 45”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GLM-5.3-Flash"; 25 input configurations explicitly selected (25 of 80 models); models slug=glm-5-3-flash; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • GPT-5.6 Sol (max)onArtificial Analysis Healthcare & Medical Index45

      “GPT-5.6 Sol (max) | 45”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GPT-5.6 Sol (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gpt-5-6-sol; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Kimi K3 (max)onArtificial Analysis Healthcare & Medical Index45

      “Kimi K3 (max) | 45”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Kimi K3 (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=kimi-k3; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Grok 4.6 (high)onArtificial Analysis Healthcare & Medical Index44

      “{"label":"Grok 4.6 (high)","score":43.92324392614704,"detailsUrl":"/models/grok-4-6"}”

      embedded JSON-LD Dataset "Artificial Analysis Healthcare & Medical Index" (Score chart data), entry 8 of 20 listed; chart header "23 of 72 models"; retrieved 2026-09-07

    • GPT-6 Sol (max)onArtificial Analysis Healthcare & Medical Index43

      “GPT-6 Sol (max) | 43”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GPT-6 Sol (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gpt-6-sol; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Gemini 3.8 Flash (high)onArtificial Analysis Healthcare & Medical Index42

      “Gemini 3.8 Flash (high) | 42”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Gemini 3.8 Flash (high)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gemini-3-8-flash; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • MiMo-V2.6-ProonArtificial Analysis Healthcare & Medical Index42

      “MiMo-V2.6-Pro | 42”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "MiMo-V2.6-Pro"; 25 input configurations explicitly selected (25 of 80 models); models slug=mimo-v2-6-pro; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • DeepSeek V4.1 Flash (Reasoning, Max Effort)onArtificial Analysis Healthcare & Medical Index41

      “DeepSeek V4.1 Flash (max) | 41”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "DeepSeek V4.1 Flash (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=deepseek-v4-1-flash; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Qwen3.8 Max (0902)onArtificial Analysis Healthcare & Medical Index41

      “Qwen3.8 Max (0902) | 41”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Qwen3.8 Max (0902)"; 25 input configurations explicitly selected (25 of 80 models); models slug=qwen3-8-max; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Step 5 PreviewonArtificial Analysis Healthcare & Medical Index41

      “Step 5 Preview | 41”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Step 5 Preview"; 25 input configurations explicitly selected (25 of 80 models); models slug=step-5; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • GPT-5.6 Terra (max)onArtificial Analysis Healthcare & Medical Index39

      “{"label":"GPT-5.6 Terra (max)","score":38.66673219960154,"detailsUrl":"/models/gpt-5-6-terra"}”

      embedded JSON-LD Dataset "Artificial Analysis Healthcare & Medical Index" (Score chart data), entry 12 of 20 listed; chart header "23 of 72 models"; retrieved 2026-09-07

    • GPT-6 Luna (max)onArtificial Analysis Healthcare & Medical Index37

      “GPT-6 Luna (max) | 37”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GPT-6 Luna (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gpt-6-luna; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • GPT-5.6 Luna (max)onArtificial Analysis Healthcare & Medical Index36

      “GPT-5.6 Luna (max) | 36”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "GPT-5.6 Luna (max)"; 25 input configurations explicitly selected (25 of 80 models); models slug=gpt-5-6-luna; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Qwen3.8 27B (xhigh)onArtificial Analysis Healthcare & Medical Index34

      “Qwen3.8 27B (xhigh) | 34”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Qwen3.8 27B (xhigh)"; 25 input configurations explicitly selected (25 of 80 models); models slug=qwen3-8-27b; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • MiniMax-M3onArtificial Analysis Healthcare & Medical Index30

      “MiniMax-M3 | 30”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "MiniMax-M3"; 25 input configurations explicitly selected (25 of 80 models); models slug=minimax-m3; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Inkling (xhigh)onArtificial Analysis Healthcare & Medical Index25

      “Inkling | 25”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Inkling"; 25 input configurations explicitly selected (25 of 80 models); models slug=inkling; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Gemini 3.5 Flash-LiteonArtificial Analysis Healthcare & Medical Index23

      “Gemini 3.5 Flash-Lite | 23”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Gemini 3.5 Flash-Lite"; 25 input configurations explicitly selected (25 of 80 models); models slug=gemini-3-5-flash-lite; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Nemotron 3 Ultra 550B A55B (Reasoning)onArtificial Analysis Healthcare & Medical Index23

      “Nemotron 3 Ultra | 23”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Nemotron 3 Ultra"; 25 input configurations explicitly selected (25 of 80 models); models slug=nvidia-nemotron-3-ultra-550b-a55b; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • Muse Glimmer (high)onArtificial Analysis Healthcare & Medical Index18

      “Muse Glimmer (high) | 18”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Muse Glimmer (high)"; 25 input configurations explicitly selected (25 of 80 models); models slug=muse-glimmer; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

    • gpt-oss-120b (high)onArtificial Analysis Healthcare & Medical Index14

      “\"slug\":\"gpt-oss-120b\",\"name\":\"gpt-oss-120b (high)\" ... \"headlineValue\":14.423032488856672”

      page data payload initialModels (RSC flight, JSON-escaped inside self.__next_f.push in the cached HTML), entry for slug "gpt-oss-120b", rank 23 of 23; the Score chart JSON-LD lists only the top 20; chart header "23 of 72 models"; retrieved 2026-09-07

    • Mistral Medium 3.5onArtificial Analysis Healthcare & Medical Index14

      “Mistral Medium 3.5 | 14”

      Live rendered Score chart, Artificial Analysis Healthcare & Medical Index, bar labelled "Mistral Medium 3.5"; 25 input configurations explicitly selected (25 of 80 models); models slug=mistral-medium-3-5; accessed 2026-09-30. Full data endpoint https://artificialanalysis.ai/data/47ff4911544c0917.txt, models[].capabilities.healthcareAndMedical.headlineValue. No board update date disclosed.

  3. Health Optimization Bench official results Official leaderboard from Health Optimization Bench. First-party.

    34 results from this document

Launch posts

  1. Introducing Grok 4.7 Launch post from SpaceXAI, published 21 September 2026. First-party.

    2 results from this document
    • Grok 4.7onHealthBench Professional56.7%

      SpaceXAI evaluation; xHigh effort; HealthBench Professional

      “Grok 4.7 xHigh Grok 4.6 High GPT-5.6 Sol Max Fable 5.1 Max Clinical reasoningHealthBench Professional 56.7% 48.5% 60.5% 62.1%”

      Model Improvements comparison table, Clinical reasoning / HealthBench Professional row; Grok 4.7 column; September 21, 2026.

    • Grok 4.6onHealthBench Professional48.5%

      SpaceXAI evaluation; High effort; HealthBench Professional

      “Grok 4.7 xHigh Grok 4.6 High GPT-5.6 Sol Max Fable 5.1 Max Clinical reasoningHealthBench Professional 56.7% 48.5% 60.5% 62.1%”

      Model Improvements comparison table, Clinical reasoning / HealthBench Professional row; Grok 4.6 column; September 21, 2026.

  2. GPT-6 Astra: A new generation of intelligence Launch post from OpenAI, published 3 September 2026, checked 8 September 2026. First-party.

    1 result from this document
    • GPT-6 AstraonHealthBench Professional63.4%

      length-adjusted, max reasoning effort (68.2 unadjusted, 3,185 mean response chars, as corrected by OpenAI on September 22; first printed as 69.5 and 4,097); GPT-6 Astra system card Table 6, column 'gpt-6 Astra'.

      This document repeats the primary source.

      “| HealthBench Professional (length-adjusted) | 63.4% | 60.5% | 58.1% | 60.9% | 56.4% | 52.1% |”

      Science and Health table

Papers

  1. Baichuan-M3 Technical Report Paper from Baichuan, published 6 February 2026, checked 7 September 2026.

    2 results from this document
    • Baichuan-M3onHealthBench65.1

      self-run in Baichuan-M3 paper (arXiv 2602.06570)

      “On the comprehensive HealthBench Total, Baichuan-M3 achieves a score of 65.1, surpassing the runner-up GPT-5.2-High (63.3) by a clear margin.”

      p. 23, section 4.2.1 HealthBench-Main; Table on p. 25 (Model / HealthBench Score)

    • GPT-5.2-HighonHealthBench63.3

      raw score as run by Baichuan in the M3 technical report, not an OpenAI-reported number; OpenAI own GPT-5.2 figure is 56.8 length-adjusted (60.7 unadjusted) in the GPT-5.6 system card.

      “GPT-5.2-High 63.3 2.37% 2.78%”

      p. 25, HealthBench-Hallu table (Model / HealthBench Score column); also p. 23 prose

Independent leaderboards

  1. HealthBench Professional Benchmark | Artificial Analysis Independent leaderboard from Artificial Analysis, published 2 November 2026, checked 2 November 2026.

    5 results from this document
    • Gemini 4 ArgononHealthBench Professional55.1%

      Artificial Analysis independent run, high reasoning; GPT-5.4 (low) judge, 4 repeats per task; length-adjusted (64.2 raw, 5,133 mean response chars)

      “HealthBench Professional: Length-Adjusted Score | Gemini 4 Argon (high) | 55.1%”

      Score section, Length-Adjusted Score chart, Gemini 4 Argon (high) bar label; raw score 64.2% on the Raw Score tab

    • Kimi K3onHealthBench Professional53.2%

      Artificial Analysis independent run, max reasoning; GPT-5.4 (low) judge, 4 repeats per task; length-adjusted (56.9 raw, 3,250 mean response chars)

      “HealthBench Professional: Length-Adjusted Score | Kimi K3 (max) | 53.2%”

      Score section, Length-Adjusted Score chart, Kimi K3 (max) bar label; raw score 56.9% on the Raw Score tab

    • GLM 5.3onHealthBench Professional52.8%

      Artificial Analysis independent run, max reasoning; GPT-5.4 (low) judge, 4 repeats per task; length-adjusted (57.9 raw, 3,748 mean response chars)

      “HealthBench Professional: Length-Adjusted Score | GLM-5.3 (max) | 52.8%”

      Score section, Length-Adjusted Score chart, GLM-5.3 (max) bar label; raw score 57.9% on the Raw Score tab

    • Gemini 3.8 FlashonHealthBench Professional51.7%

      Artificial Analysis independent run, high reasoning; GPT-5.4 (low) judge, 4 repeats per task; length-adjusted (59.0 raw, 4,494 mean response chars)

      “HealthBench Professional: Length-Adjusted Score | Gemini 3.8 Flash (high) | 51.7%”

      Score section, Length-Adjusted Score chart, Gemini 3.8 Flash (high) bar label; raw score 59.0% on the Raw Score tab

    • DeepSeek V4.1 FlashonHealthBench Professional50.7%

      Artificial Analysis independent run, max reasoning; GPT-5.4 (low) judge, 4 repeats per task; length-adjusted (52.9 raw, 2,760 mean response chars)

      “HealthBench Professional: Length-Adjusted Score | DeepSeek V4.1 Flash (max) | 50.7%”

      Score section, Length-Adjusted Score chart, DeepSeek V4.1 Flash (max) bar label; raw score 52.9% on the Raw Score tab