HealthBench Professional — Mercor evaluation (HealthBench Professional (Mercor))

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

Answering clinicians in care consults, documentation and medical research conversations. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Evaluation results published by Mercor. Published evaluation aggregates only; task contents and dataset license grants are not included.

Adjusted Score on HealthBench Professional (Mercor) — November 12, 2026 capture

We mirror the published adjusted score view for HealthBench Professional (Mercor). GPT-6 Astra leads the public snapshot at 64.50%, followed by GPT-6.1 Sol (64.40%) and Sonnet 5.5 (62.70%). We do not use these results to rank models overall.

GPT-6 Astra

OpenAI

Single-shot agent · max reasoning

Margin not reported · 2,100 reported samples

64.50%

GPT-6.1 Sol

OpenAI

Single-shot agent · max reasoning

Margin not reported · 2,100 reported samples

64.40%

Sonnet 5.5

Anthropic

Single-shot agent · max reasoning

Margin not reported · 2,100 reported samples

62.70%

37 configurationsKnowledgeDated Mercor configurationsDisplay onlyUpdated November 12, 2026 capture

Adjusted Score table (37 configurations)

Adjusted Score results for HealthBench Professional (Mercor)
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	GPT-6 AstraOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	64.50%	Not reported	Unknown
2	GPT-6.1 SolOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	64.40%	Not reported	Unknown
3	Sonnet 5.5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	62.70%	Not reported	Unknown
4	GPT-6 LunaOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	62.20%	Not reported	Unknown
5	GPT-6 SolOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	61.40%	Not reported	Unknown
6	GPT-5.6 Sol (Pro)OpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	61.40%	Not reported	Unknown
7	Opus 5.5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	60.40%	Not reported	Unknown
8	Fable 5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	59.00%	Not reported	Unknown
9	Haiku 5.5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	58.80%	Not reported	Unknown
10	Muse Spark OpenMetaSingle-shot agent · xhigh reasoningMargin not reported · 1,050 reported samples	57.20%	Not reported	Unknown
11	Fable 5.1AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	56.50%	Not reported	Unknown
12	GPT-5.6 TerraOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	56.30%	Not reported	Unknown
13	Grok 4.7xAISingle-shot agent · xhigh reasoningMargin not reported · 2,100 reported samples	56.20%	Not reported	Unknown
14	GPT-5.6 LunaOpenAISingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	55.90%	Not reported	Unknown
15	Opus 5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	55.40%	Not reported	Unknown
16	Gemini 4 ArgonGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	55.40%	Not reported	Unknown
17	GLM-5.3ZhipuSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	52.90%	Not reported	Unknown
18	Sonnet 5AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	52.50%	Not reported	Unknown
19	Gemini 3.8 FlashGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	52.40%	Not reported	Unknown
20	GPT-5.5OpenAISingle-shot agent · xhigh reasoningMargin not reported · 2,100 reported samples	51.20%	Not reported	Unknown
21	Kimi K3KimiSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	50.20%	Not reported	Unknown
22	DeepSeek-V4.1-FlashDeepSeekSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	49.90%	Not reported	Unknown
23	Opus 4.7AnthropicSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	49.70%	Not reported	Unknown
24	Gemini 3.7 FlashGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	48.40%	Not reported	Unknown
25	Gemini 3.6 FlashGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	46.40%	Not reported	Unknown
26	Grok 4.6xAISingle-shot agent · xhigh reasoningMargin not reported · 2,100 reported samples	44.40%	Not reported	Unknown
27	Gemini 3.1 ProGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	43.90%	Not reported	Unknown
28	Gemini 3.5 FlashGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	43.40%	Not reported	Unknown
29	Grok 4.5xAISingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	42.50%	Not reported	Unknown
30	BeamReflectionSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	41.70%	Not reported	Unknown
31	InklingThinking MachinesSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	41.20%	Not reported	Unknown
32	DeepSeek-V4-Pro-0813DeepSeekSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	40.90%	Not reported	Unknown
33	Qwen3.8-MaxAlibabaSingle-shot agent · xhigh reasoningMargin not reported · 2,100 reported samples	39.00%	Not reported	Unknown
34	MiniMax-M3MiniMaxSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	37.10%	Not reported	Unknown
35	Gemini 3.5 Flash LiteGoogleSingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	35.00%	Not reported	Unknown
36	Nemotron 3 UltraNVIDIASingle-shot agent · high reasoningMargin not reported · 2,100 reported samples	32.90%	Not reported	Unknown
37	DeepSeek-V4-FlashDeepSeekSingle-shot agent · max reasoningMargin not reported · 2,100 reported samples	32.80%	Not reported	Unknown

How to read this Mercor evaluation

We captured Mercor's HealthBench Professional results on November 12, 2026. The table preserves 37 published configurations, their source identifiers, reasoning settings, and disclosed adjusted score values. These are results from Mercor's evaluation setup; we did not rerun them.

This dated source table is display only and does not enter overall or category rankings. Source configurations retain their external identity, including Pro descriptors and different effort settings.

The headline is the adjusted score Mercor publishes for the 525 physician-authored tasks. Margins are not reported. Scores labs report on their own runs, and other evaluators' runs, are listed separately and are not combined with this table.

The row labels retain published error margins and sample counts when supplied. Missing margins and counts remain unreported; a sample count is not used to reconstruct a task denominator. Except where the dedicated source defines an interval, the confidence level is unreported. Equal displayed scores and overlapping margins do not establish statistical ties. Capture time does not establish individual evaluation dates.

The general methodology lists expert-authored rubrics, scored criterion by criterion and a judge of LLM judge over expert rubrics. Its run-count label is k = 4. The linked result page takes precedence where its setup differs, as noted above.

Snapshot

Adjusted Score37 configurationsMercor evaluationDisplay only

Mercor HealthBench Professional results Mercor APEX benchmark index Mercor methodology Source paper Upstream evaluation code Upstream dataset

The published HealthBench Professional (Mercor) snapshot places GPT-6 Astra first at 64.50%. The third row is 1.80 points behind. The broader top-10 range is 7.30 points, so many of the published results sit in a relatively narrow band.

37 configurations are shown for HealthBench Professional (Mercor). The benchmark falls in the Knowledge category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. HealthBench Professional (Mercor) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About HealthBench Professional (Mercor)

Tasks

525 clinician tasks

Format

Adjusted Score

Difficulty

Source-specific evaluation

Mercor runs the 525 physician-authored HealthBench Professional tasks and publishes the adjusted score. Scores the labs report themselves stay on the HealthBench Professional page.

HealthBench Professional evaluation results Public benchmark source

Freshness and provenance

Version

HealthBench Professional; Mercor Adjusted Score

Refresh cadence

Reviewed dated source capture

Staleness state

Dated Mercor configurations

Question availability

Upstream task access and terms; only aggregate results mirrored

Dated Mercor configurationsDisplay 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 HealthBench Professional (Mercor) measure?

Answering clinicians in care consults, documentation and medical research conversations. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Which model leads the published HealthBench Professional (Mercor) snapshot?

GPT-6 Astra currently leads the published HealthBench Professional (Mercor) snapshot with 64.50% adjusted score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on HealthBench Professional (Mercor)?

The November 12, 2026 capture snapshot contains 37 source configurations.

Last updated: November 12, 2026 capture · 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 7,000+ subscribers.

One email each week. Unsubscribe anytime.