SAFE-Teen (Vals AI)
Does the model respond safely when the stakes are personal?
More about this benchmark
How often chatbots fail a teen who hints at risk or turns down help; lower is better.
Whether chatbots stay safe when a teen (13 to 17) brings up a mental health concern: a simulated teen talks with each model for 6 to 20 turns, branching when the teen accepts or refuses help; a judge rates each conversation against a clinician rubric. Headline is the urgent-safety failure rate (suicide, self-harm, abuse, violence, eating disorders, psychosis, substance use) without a teen-aware system prompt.
- On this site
- 17 of 18 models
- Released
- Oct 2026
- Headline metric
- Urgent safety failure rate, no teen-aware prompt
- Official board updated
- 13 Oct 2026
Scores come from runs by the benchmark's maintainers or independent researchers.
- Results
- 18 from 17 models
- Latest
- Oct 2026
- Published by
- Vals AI
- Index weight
- 0.131, narrow, near its ceiling
- Sources
- 1
Counts toward the Health Evals Index as a narrow benchmark, at weight 0.131: teen mental-health safety, one population. It is near its ceiling, with a leader at 85.0 out of 100, so it counts for less.
Where every published SAFE-Teen (Vals AI) score sits
18 scores, one column each, highest first, on the benchmark's own scale. Lower is better.
Each column is the model's lab colour. The table below says who ran each score.
Every published score, ranked within this benchmark only
18 scores. Lower is better. Where the source gave a range, the whisker shows it. Open any row to read the sentence the score came from.
1GPT-6.1 SolOpenAI · no teen-aware prompt15Benchmark teamOct 2026[1]
GPT-6.1 Sol | 15% | 9% | 19% (Urgent safety | Health advice | Relationship boundaries)
2Claude Sonnet 5.5Anthropic · no teen-aware prompt23Benchmark teamOct 2026[1]
Claude Sonnet 5.5 | 23% | 43% | 37% (Urgent safety | Health advice | Relationship boundaries)
2GPT-5.6 LunaOpenAI · no teen-aware prompt23Benchmark teamOct 2026[1]
GPT-5.6 Luna | 23% | 21% | 35% (Urgent safety | Health advice | Relationship boundaries)
4Muse Spark 1.3 (Extra High)Meta · no teen-aware prompt26Benchmark teamOct 2026[1]
Muse Spark 1.3 (Extra High) | 26% | 14% | 56% (Urgent safety | Health advice | Relationship boundaries)
4Qwen 3.8 MaxAlibaba · no teen-aware prompt26Benchmark teamOct 2026[1]
Qwen 3.8 Max | 26% | 35% | 51% (Urgent safety | Health advice | Relationship boundaries)
6Claude Opus 5.5Anthropic · no teen-aware prompt27Benchmark teamOct 2026[1]
Claude Opus 5.5 | 27% | 44% | 25% (Urgent safety | Health advice | Relationship boundaries)
7DeepSeek V4.1 FlashDeepSeek · no teen-aware prompt31Benchmark teamOct 2026[1]
DeepSeek V4.1 Flash | 31% | 41% | 51% (Urgent safety | Health advice | Relationship boundaries)
8Mistral Large 4 PreviewMistral · no teen-aware prompt33Benchmark teamOct 2026[1]
Mistral Large 4 Preview | 33% | 53% | 19% (Urgent safety | Health advice | Relationship boundaries)
8Muse Spark 1.3 (Minimal)Meta · no teen-aware prompt33Benchmark teamOct 2026[1]
Muse Spark 1.3 (Minimal) | 33% | 29% | 70% (Urgent safety | Health advice | Relationship boundaries)
10MiMo v2.6 ProXiaomi · no teen-aware prompt40Benchmark teamOct 2026[1]
MiMo v2.6 Pro | 40% | 57% | 56% (Urgent safety | Health advice | Relationship boundaries)
11Grok 4.7SpaceX AI · no teen-aware prompt44Benchmark teamOct 2026[1]
Grok 4.7 | 44% | 39% | 48% (Urgent safety | Health advice | Relationship boundaries)
11Kimi K3Moonshot AI · no teen-aware prompt44Benchmark teamOct 2026[1]
Kimi K3 | 44% | 61% | 63% (Urgent safety | Health advice | Relationship boundaries)
13Grok 4.5SpaceX AI · no teen-aware prompt50Benchmark teamOct 2026[1]
Grok 4.5 | 50% | 47% | 65% (Urgent safety | Health advice | Relationship boundaries)
14GLM 5.3Zhipu · no teen-aware prompt51Benchmark teamOct 2026[1]
GLM 5.3 | 51% | 70% | 64% (Urgent safety | Health advice | Relationship boundaries)
15Gemini 3.6 FlashGoogle · no teen-aware prompt52Benchmark teamOct 2026[1]
Gemini 3.6 Flash | 52% | 68% | 70% (Urgent safety | Health advice | Relationship boundaries)
16Gemini 4 ArgonGoogle · no teen-aware prompt53Benchmark teamOct 2026[1]
Gemini 4 Argon | 53% | 71% | 66% (Urgent safety | Health advice | Relationship boundaries)
17MiniMax M3MiniMax · no teen-aware prompt54Benchmark teamOct 2026[1]
MiniMax M3 | 54% | 49% | 53% (Urgent safety | Health advice | Relationship boundaries)
18Gemini 3.8 FlashGoogle · no teen-aware prompt56Benchmark teamOct 2026[1]
Gemini 3.8 Flash | 56% | 69% | 63% (Urgent safety | Health advice | Relationship boundaries)
Health advice failure rate, ranked on its own
18 scores. Lower is better. Where the source gave a range, the whisker shows it. Open any row to read the sentence the score came from.
1GPT-6.1 SolOpenAI · no teen-aware prompt9Benchmark teamOct 2026[1]
GPT-6.1 Sol | 15% | 9% | 19%
2Muse Spark 1.3 (Extra High)Meta · no teen-aware prompt14Benchmark teamOct 2026[1]
Muse Spark 1.3 (Extra High) | 26% | 14% | 56%
3GPT-5.6 LunaOpenAI · no teen-aware prompt21Benchmark teamOct 2026[1]
GPT-5.6 Luna | 23% | 21% | 35%
4Muse Spark 1.3 (Minimal)Meta · no teen-aware prompt29Benchmark teamOct 2026[1]
Muse Spark 1.3 (Minimal) | 33% | 29% | 70%
5Qwen 3.8 MaxAlibaba · no teen-aware prompt35Benchmark teamOct 2026[1]
Qwen 3.8 Max | 26% | 35% | 51%
6Grok 4.7SpaceX AI · no teen-aware prompt39Benchmark teamOct 2026[1]
Grok 4.7 | 44% | 39% | 48%
7DeepSeek V4.1 FlashDeepSeek · no teen-aware prompt41Benchmark teamOct 2026[1]
DeepSeek V4.1 Flash | 31% | 41% | 51%
8Claude Sonnet 5.5Anthropic · no teen-aware prompt43Benchmark teamOct 2026[1]
Claude Sonnet 5.5 | 23% | 43% | 37%
9Claude Opus 5.5Anthropic · no teen-aware prompt44Benchmark teamOct 2026[1]
Claude Opus 5.5 | 27% | 44% | 25%
10Grok 4.5SpaceX AI · no teen-aware prompt47Benchmark teamOct 2026[1]
Grok 4.5 | 50% | 47% | 65%
11MiniMax M3MiniMax · no teen-aware prompt49Benchmark teamOct 2026[1]
MiniMax M3 | 54% | 49% | 53%
12Mistral Large 4 PreviewMistral · no teen-aware prompt53Benchmark teamOct 2026[1]
Mistral Large 4 Preview | 33% | 53% | 19%
13MiMo v2.6 ProXiaomi · no teen-aware prompt57Benchmark teamOct 2026[1]
MiMo v2.6 Pro | 40% | 57% | 56%
14Kimi K3Moonshot AI · no teen-aware prompt61Benchmark teamOct 2026[1]
Kimi K3 | 44% | 61% | 63%
15Gemini 3.6 FlashGoogle · no teen-aware prompt68Benchmark teamOct 2026[1]
Gemini 3.6 Flash | 52% | 68% | 70%
16Gemini 3.8 FlashGoogle · no teen-aware prompt69Benchmark teamOct 2026[1]
Gemini 3.8 Flash | 56% | 69% | 63%
17GLM 5.3Zhipu · no teen-aware prompt70Benchmark teamOct 2026[1]
GLM 5.3 | 51% | 70% | 64%
18Gemini 4 ArgonGoogle · no teen-aware prompt71Benchmark teamOct 2026[1]
Gemini 4 Argon | 53% | 71% | 66%
Relationship boundaries failure rate, ranked on its own
18 scores. Lower is better. Where the source gave a range, the whisker shows it. Open any row to read the sentence the score came from.
1GPT-6.1 SolOpenAI · no teen-aware prompt19Benchmark teamOct 2026[1]
GPT-6.1 Sol | 15% | 9% | 19%
1Mistral Large 4 PreviewMistral · no teen-aware prompt19Benchmark teamOct 2026[1]
Mistral Large 4 Preview | 33% | 53% | 19%
3Claude Opus 5.5Anthropic · no teen-aware prompt25Benchmark teamOct 2026[1]
Claude Opus 5.5 | 27% | 44% | 25%
4GPT-5.6 LunaOpenAI · no teen-aware prompt35Benchmark teamOct 2026[1]
GPT-5.6 Luna | 23% | 21% | 35%
5Claude Sonnet 5.5Anthropic · no teen-aware prompt37Benchmark teamOct 2026[1]
Claude Sonnet 5.5 | 23% | 43% | 37%
6Grok 4.7SpaceX AI · no teen-aware prompt48Benchmark teamOct 2026[1]
Grok 4.7 | 44% | 39% | 48%
7DeepSeek V4.1 FlashDeepSeek · no teen-aware prompt51Benchmark teamOct 2026[1]
DeepSeek V4.1 Flash | 31% | 41% | 51%
7Qwen 3.8 MaxAlibaba · no teen-aware prompt51Benchmark teamOct 2026[1]
Qwen 3.8 Max | 26% | 35% | 51%
9MiniMax M3MiniMax · no teen-aware prompt53Benchmark teamOct 2026[1]
MiniMax M3 | 54% | 49% | 53%
10MiMo v2.6 ProXiaomi · no teen-aware prompt56Benchmark teamOct 2026[1]
MiMo v2.6 Pro | 40% | 57% | 56%
10Muse Spark 1.3 (Extra High)Meta · no teen-aware prompt56Benchmark teamOct 2026[1]
Muse Spark 1.3 (Extra High) | 26% | 14% | 56%
12Gemini 3.8 FlashGoogle · no teen-aware prompt63Benchmark teamOct 2026[1]
Gemini 3.8 Flash | 56% | 69% | 63%
12Kimi K3Moonshot AI · no teen-aware prompt63Benchmark teamOct 2026[1]
Kimi K3 | 44% | 61% | 63%
14GLM 5.3Zhipu · no teen-aware prompt64Benchmark teamOct 2026[1]
GLM 5.3 | 51% | 70% | 64%
15Grok 4.5SpaceX AI · no teen-aware prompt65Benchmark teamOct 2026[1]
Grok 4.5 | 50% | 47% | 65%
16Gemini 4 ArgonGoogle · no teen-aware prompt66Benchmark teamOct 2026[1]
Gemini 4 Argon | 53% | 71% | 66%
17Gemini 3.6 FlashGoogle · no teen-aware prompt70Benchmark teamOct 2026[1]
Gemini 3.6 Flash | 52% | 68% | 70%
17Muse Spark 1.3 (Minimal)Meta · no teen-aware prompt70Benchmark teamOct 2026[1]
Muse Spark 1.3 (Minimal) | 33% | 29% | 70%
The 1 document behind these scores
- 1.SAFE-Teen, Vals AIofficial leaderboard, Vals AI, 13 Oct 2026. Quotes from this document