Skip to main content
BenchLM
Data

Vals Mental Health Bench (Mental Health Bench (Vals))

We show this table for reference; we do not rank on it.

Data verified

Published chatbot failure rates across urgent safety, health advice, and relationship boundaries in simulated teen conversations.

Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.

Urgent safety failure rate on Mental Health Bench (Vals) — Captured October 15, 2026

We mirror the published urgent safety failure rate view for Mental Health Bench (Vals). The public snapshot contains 18 models. GPT-6.1 Sol has the lowest published score at 15%, but the available coverage and evaluation setups do not establish a market leader. We do not use these results to rank models overall.

1Closed

GPT-6.1 Sol

OpenAI

15%
Overall 84.07Context 1.05M
2Closed

GPT-5.6 Luna

OpenAI

23%
Overall 64.35Context 1.05M
3Closed

Claude Sonnet 5.5

Anthropic

23%
Overall 82.18Context 1M

18 modelsCross-domainCurrentDisplay onlyUpdated Captured October 15, 2026

Urgent safety failure rate table (18 models)

Urgent safety failure rate results for Mental Health Bench (Vals)
RankModel / configurationScoreParameters (B)Open / closed
1GPT-6.1 SolOpenAI
15%
Not reportedClosed
2GPT-5.6 LunaOpenAI
23%
Not reportedClosed
3Claude Sonnet 5.5Anthropic
23%
Not reportedClosed
4Muse Spark 1.3 (Extra High)Unknown
26%
Not reportedUnknown
5Qwen 3.8 MaxAlibaba
26%
Not reportedUnknown
6Claude Opus 5.5Anthropic
27%
Not reportedClosed
7DeepSeek V4.1 Flash 1017DeepSeek
31%
Not reportedOpen
8Mistral Large 4Mistral
33%
Not reportedPending
9Muse Spark 1.3 (Minimal)Unknown
33%
Not reportedUnknown
10MiMo-V2.6-ProXiaomi
40%
Not reportedOpen
11Kimi K3Moonshot AI
44%
Not reportedPending
12Grok 4.7xAI
44%
Not reportedClosed
13Grok 4.5xAI
50%
Not reportedClosed
14GLM-5.3Z.AI
51%
Not reportedOpen
15Gemini 3.6 FlashGoogle
52%
Not reportedClosed
16Gemini 4 ArgonGoogle
53%
Not reportedClosed
17MiniMax M3MiniMax
54%
Not reportedOpen
18Gemini 3.8 FlashGoogle
56%
Not reportedClosed

How Mental Health Bench is shown here

We mirror Vals AI's published failure rates for simulated teen conversations. The table is ordered by urgent safety failure rate; lower is better. Health advice and relationship-boundary failure rates remain separate fields in the source snapshot.

The percentages are rounded as published. They measure this evaluation's conversations and judge settings, not clinical outcomes or a guarantee that a chatbot is safe. These results are display only and do not enter model rankings.

Snapshot

18 Vals rows3 separate risk areasLower failure rate is betterRounded source percentagesDisplay only
Mental Health Bench on Vals AIVals methodologyVals home

The published Mental Health Bench (Vals) snapshot places GPT-6.1 Sol first at 15%. The third row is 8 points higher. The broader top-10 range is 25 points, so the table still separates the published systems.

18 models have been evaluated on Mental Health Bench (Vals). The benchmark falls in the Cross-domain category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Mental Health Bench (Vals) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Mental Health Bench (Vals)

Year

2026

Tasks

Simulated teen conversations across three risk areas

Format

Urgent safety failure rate

Difficulty

Conversation safety evaluation

The table shows the source's rounded urgent safety failure rate, with lower values indicating fewer failures in this evaluation. Health advice and relationship-boundary rates remain separate in the snapshot. These results are display only, do not measure clinical outcomes, and do not enter model rankings.

Mental Health BenchPublic benchmark source

Freshness and provenance

Version

Mental Health Bench (Vals) 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does Mental Health Bench (Vals) measure?

Published chatbot failure rates across urgent safety, health advice, and relationship boundaries in simulated teen conversations.

Which model leads the published Mental Health Bench (Vals) snapshot?

GPT-6.1 Sol currently leads the published Mental Health Bench (Vals) snapshot with 15% urgent safety failure rate. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Mental Health Bench (Vals)?

The Captured October 15, 2026 snapshot contains 18 AI models.

Last updated: Captured October 15, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 5,600+ readers.

One email each week. Unsubscribe anytime.