Benchmark Almanac

MentalHealthBench

GPT-6 Astra 57.3% leads since September 2026
GPT-6.1 Sol 57.9% leads since September 2026
Only in system cards

The benchmark’s paper hasn’t scored a model released after GPT-6 Sol (September 2026). Labs report their own numbers for 4 newer models, such as GPT-6.1 Sol at 57.9%.

1,215 synthetic mental-health conversations with 5,262 clinician-authored rubric criteria. An LLM judge scores safety and guidance quality.

Grader is GPT-5.6 Sol. More than 80 licensed psychologists and psychiatrists from more than 20 countries wrote and adjudicated the criteria. The paper URL for the data (no extension) returns 404, but OAI_MentalHealthBench.zip resolves.

Made by
OpenAI
Released
September 2026
Kind
Wellbeing and sycophancy
Level
Expert
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Multi-turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a system card
November 2026
Last model added
November 2026
Scores refreshed
November 18th, 2026
17
Entries in its paper, 12 models we track
7
System cards report it, from 1 lab

GPT-6 Astra leads with 57.3% among models released by September 2026 GPT-6 Astra leads with 57.3% among models released by September 2026

Best score per entry from the benchmark's paper, read on November 18th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

GPT-6.1 Sol leads with 57.9%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model. Striped bars are claims, and dotted bars are reruns by Epoch AI or another board.

The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-6 Astra default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 57.3%
  2. 2 GPT-6 Sol default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 53.9%
  3. 3 Claude Opus 5.5 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 52.4%
  4. 4 GPT-6 Luna default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 50.2%
  5. 5 Muse Spark 1.3 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 48.6%
  6. 6 GPT-5.6 Sol (Aug 2026) default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 47%
  7. 7 Claude Fable 5.1 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 46.4%
  8. 8 GPT-5.6 Luna (Aug 2026) default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 44.9%
  9. 9 Claude Sonnet 5 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 44.5%
  10. 10 GPT-5 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 42.9%
  11. 11 Claude Haiku 4.5 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 41.7%
  12. 12 Grok 4.7 default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 41.3%
  13. 13 Gemini 3.8 Flash default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 35.5%
  14. 14 Gemini 2.5 Flash (snapshot unspecified) default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 33.5%
  15. 15 GPT-4o (March 2025) default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 32.1%
  16. 16 Gemini 3.1 Pro default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 32.1%
  17. 17 Gemini 2.5 Pro (snapshot unspecified) default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge 29.5%

Source: The MentalHealthBench paper

Source: The MentalHealthBench paper

Sources: Labs’ model cards and the MentalHealthBench paper

#EntryScoreSettingDateSource
1 GPT-6 Astra 57.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
2 GPT-6 Sol 53.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
3 Claude Opus 5.5 52.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
4 GPT-6 Luna 50.2% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
5 Muse Spark 1.3 48.6% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
6 GPT-5.6 Sol (Aug 2026) 47% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Official leaderboard
7 Claude Fable 5.1 46.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
8 GPT-5.6 Luna (Aug 2026) 44.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Official leaderboard
9 Claude Sonnet 5 44.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Jun 2026 Official leaderboard
10 GPT-5 42.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Aug 2025 Official leaderboard
11 Claude Haiku 4.5 41.7% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Oct 2025 Official leaderboard
12 Grok 4.7 41.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
13 Gemini 3.8 Flash 35.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
14 Gemini 2.5 Flash (snapshot unspecified) 33.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Official leaderboard
15 GPT-4o (March 2025) 32.1% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Official leaderboard
16 Gemini 3.1 Pro 32.1% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Feb 2026 Official leaderboard
17 Gemini 2.5 Pro (snapshot unspecified) 29.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6 Astra 57.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
2 GPT-6 Sol 53.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
3 Claude Opus 5.5 52.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
4 GPT-6 Luna 50.2% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
5 Muse Spark 1.3 48.6% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
6 Claude Fable 5.1 46.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
7 Claude Sonnet 5 44.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Jun 2026 Official leaderboard
8 GPT-5 42.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Aug 2025 Official leaderboard
9 Claude Haiku 4.5 41.7% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Oct 2025 Official leaderboard
10 Grok 4.7 41.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
11 Gemini 3.8 Flash 35.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
12 Gemini 3.1 Pro 32.1% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Feb 2026 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-6.1 Sol 57.9% max effort Sep 2026 Lab model card
2 GPT-6 Astra 57.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
3 GPT-6.1 Terra 55.5% max effort Oct 2026 Lab model card
4 GPT-6 Sol 53.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
5 GPT-6.2 Astra 53.3% max effort Nov 2026 Lab model card
6 Claude Opus 5.5 52.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
7 GPT-6 Luna 51.7% Oct 2026 Lab model card
8 GPT-6.2 Sol 50.9% max effort Nov 2026 Lab model card
9 Muse Spark 1.3 48.6% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
10 Claude Fable 5.1 46.4% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
11 Claude Sonnet 5 44.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Jun 2026 Official leaderboard
12 GPT-5 42.9% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Aug 2025 Official leaderboard
13 Claude Haiku 4.5 41.7% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Oct 2025 Official leaderboard
14 Grok 4.7 41.3% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
15 Gemini 3.8 Flash 35.5% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Sep 2026 Official leaderboard
16 Gemini 3.1 Pro 32.1% default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge Feb 2026 Official leaderboard

Labs report scores for 6 models in their cards. Switch the scores above to rank them together.

Rankings by part

We read 8 parts from the benchmark’s paper in 3 groups: acuity, user and context. Each ranks models on one part of the test. GPT-6 Astra leads 6 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Non-acute

Low-urgency, everyday conversations with some emotional weight, such as venting or relationship advice. About 54% of tasks. 12 models we track have a score here.

  1. 1 GPT-6 Astra 56.6%
  2. 2 GPT-6 Sol 52.3%
  3. 3 Claude Opus 5.5 51.4%
  4. 4 GPT-6 Luna 47.9%
  5. 5 Claude Fable 5.1 ↑1 44.2%
  6. 6 Claude Sonnet 5 ↑1 43.9%
  7. 7 Muse Spark 1.3 ↓2 43.2%
  8. 8 Claude Haiku 4.5 ↑1 41.4%
  9. 9 Grok 4.7 ↑1 40.2%
  10. 10 GPT-5 ↓2 39.7%

3 models have held number one in the benchmark’s paper since its release in September 2026 4 models have held number one with claims included since its release in September 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s paper and what labs claim in their cards.

Anthropic Meta OpenAI
Benchmark paper Not tested since Sep 2026
GPT-6 Astra 57.3%
Lab claims 1 leader
GPT-6.1 Sol 57.9%

Sources: The MentalHealthBench paper and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6.1 Sol Muse Spark 1.3 GPT-5 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) GPT-6.1 Sol Muse Spark 1.3

Sources: Official MentalHealthBench leaderboard and labs’ model cards

Labs report it only in system cards

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark. Some system cards do, below.

7 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests. OpenAI accounts for 6 of the 6 releases that named it since 2024.

DateDocumentScoreAs the document writes itSection
Nov 17, 2026 GPT-6.2 Sol system card 50.9 MentalHealthBench (overall), maximum reasoning effort, 50.9 ± 1.0 (Table 19) Health
Nov 17, 2026 GPT-6.2 Astra system card 53.3 MentalHealthBench (overall), maximum reasoning effort, 53.3 ± 1.0 (Table 19) Health
Oct 29, 2026 GPT-6 Luna system card addendum 48.7 MentalHealthBench (overall), task-clipped metric (Table 6) Health
Oct 26, 2026 GPT-6.1 Terra system card addendum 55.5 MentalHealthBench (overall), maximum reasoning effort, 55.5 ± 1.0 (5.3) Health
Oct 7, 2026 GPT-6 Sol system card addendum 48.8 MentalHealthBench (overall), task-clipped metric (Table 6) Health
Oct 7, 2026 GPT-6 Luna system card addendum 51.7 MentalHealthBench (overall), task-clipped metric (Table 6) Health
Sep 29, 2026 GPT-6.1 Sol system card addendum 57.9 MentalHealthBench (overall), maximum reasoning effort (5.3, table under Figure 4) Health

The lab's number comes out ahead in 1 of 2 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
GPT-6 Sol 48.8% MentalHealthBench (overall), task-clipped metric (Table 6) 53.9% Leaderboard, default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge −5.1 pts
GPT-6 Luna 51.7% MentalHealthBench (overall), task-clipped metric (Table 6) 50.2% Leaderboard, default API reasoning, temperature and verbosity; GPT-5.6 Sol high-reasoning judge +1.5 pts

More in Safety and alignment

All 636 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
  • Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior. Inkling 2 Mini 8.30%
  • Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions. Inkling 2 Small 4.70%
  • Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior. Inkling 2 Mini 9.80%

Data audit

11 of 12 checks pass, 1 warning. Look first at the Confidence check. Likely. Some facts rest on secondary sources. Every benchmark

Catalog record 1 warning 1 pass
  • Pass Task access All tasks are public.
  • Warning Confidence Likely. Some facts rest on secondary sources.
Scores 5 pass
  • Pass Independent runs Scores from the benchmark’s paper (17 entries) and system cards (7 scores).
  • Pass Models scored 16 models we track, plus 5 other entries on its leaderboard.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Pass Frontier models 6 of the top 10 have a score, led by GPT-6.2 Astra (#1).
  • Pass Headroom The best is 57.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on cdn.openai.com: 12 rows for models we track and 5 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 2 pass
  • Pass Lab claims Both pairs agree within 10 points.
  • Pass Known issues No published audit found problems.

How we check