Clinical Benchmarks

Clinical Benchmarks Index

Claude Sonnet 5.5 is first at 63.2. Claude Opus 5.5 follows at 62.8, then Claude Opus 5 at 56.8.

This site's own calculation from 12 of 14 clinical boards, each read on its own 0 to 100 scale. Method of 31 October 2026.

Models ranked
128
Results
472
Source documents
92

Ranking, model by model

The marks under each column show which of the 12 index boards the score uses.

Select a column to open the model. The filter hides models and leaves every rank as it is.

Each index board, model by model

The top 10 models on each of the 12 index boards, on that board's own 0 to 100 scale. Each column opens its source.

HealthBench Professional

  1. 0.703GPT-6 Astra
  2. 0.692Claude Sonnet 5.5
  3. 0.660Claude Fable 5
  4. 0.656Claude Opus 5.5
  5. 0.648Claude Haiku 5.5
  6. 0.642GPT-6.1 Sol
  7. 0.627GPT-6.1 Terra
  8. 0.623Muse Wonder Full
  9. 0.621Claude Fable 5.1
  10. 0.608GPT-6 Luna

37 modelsFull board

MedScribe (Vals AI)near ceiling

  1. 91.43Claude Opus 5.5
  2. 91.29Claude Fable 5.1
  3. 91.10Claude Sonnet 5.5
  4. 90.98Claude Opus 5
  5. 90.06Muse Spark 1.2
  6. 89.38Grok 4.7
  7. 88.89Muse Spark 1.1
  8. 88.94GLM 5.3 Flash
  9. 88.81GLM 5.3
  10. 88.52Claude Fable 5

95 modelsFull board

MedCode (Vals AI)

  1. 63.57Claude Opus 5
  2. 59.06Gemini 3.1 Pro
  3. 58.80Gemini 4 Argon
  4. 56.07Claude Fable 5
  5. 55.92Gemini 3 Flash
  6. 55.83Gemini 3.5 Flash
  7. 54.86Claude Opus 4.7
  8. 53.51Claude Fable 5.1
  9. 53.39Gemini 3.7 Flash
  10. 53.22Claude Opus 4.8

95 modelsFull board

MedXpertQA (MM)near ceiling

  1. 81.5GPT-5.6 Sol
  2. 81.3Gemini 3.1 Pro
  3. 80.4Qwen3.8 Max
  4. 80.0Claude Fable 5
  5. 78.4Muse Spark
  6. 77.1GPT-5.4
  7. 76.0Gemini 3 Pro
  8. 73.3GPT-5.2
  9. 71.7Claude Opus 4.8
  10. 71.0Qwen3.7 Plus

22 modelsFull board

MAST (Medical AI Superintelligence Test)

  1. 60.2GPT-5.6 Sol
  2. 60.1Kimi K3
  3. 59.3Gemini 3.6 Flash
  4. 58.9Gemini 3.1 Pro
  5. 57.9Qwen3.5 397B A17B
  6. 57.1Claude Opus 5
  7. 56.6Claude Sonnet 5
  8. 53.7Grok 4.3

8 modelsFull board

First, Do NOHARM (v2)

  1. 79.7Muse Spark 1.1
  2. 74.6Claude Opus 5
  3. 74.0Kimi K3
  4. 70.1GPT-5.6 Sol
  5. 70.0GPT-5.5
  6. 68.6GPT-5
  7. 65.0Claude Fable 5
  8. 62.6Gemini 3.1 Pro
  9. 61.9Gemini 2.5 Pro
  10. 61.1Qwen3.5 397B A17B

13 modelsFull board

PhysicianBench

  1. 68.4Claude Opus 5.5
  2. 63.2Claude Sonnet 5.5
  3. 61.0Claude Fable 5.1
  4. 57.6Claude Opus 5
  5. 46.3GPT-5.5
  6. 43.0Claude Haiku 5.5
  7. 37.4Claude Sonnet 5
  8. 31.7Claude Opus 4.6
  9. 29.3Claude Opus 4.7
  10. 27.7GPT-5.4

19 modelsFull board

Bedside Benchnear ceiling

  1. 94.6Claude Opus 5
  2. 94.3Muse Spark Open
  3. 93.4GPT-6 Astra
  4. 92.7Gemini 4 Argon
  5. 91.1GPT-5.6 Sol
  6. 89.7Kimi K3
  7. 89.1Ember-1
  8. 88.1GPT-6 Sol
  9. 87.2GPT-6 Luna
  10. 86.9GLM 5.3

15 modelsFull board

RCM-Benchnear ceiling

  1. 92.68Claude Opus 5
  2. 91.49Gemini 4 Argon
  3. 91.21GPT-5.6 Sol
  4. 91.24Claude Sonnet 5
  5. 91.05Claude Sonnet 5.5
  6. 90.23Gemini 3.8 Flash
  7. 90.14Claude Opus 5.5
  8. 89.82Claude Sonnet 4.6
  9. 88.99DeepSeek V4.1 Flash
  10. 88.38Claude Haiku 5.5

20 modelsFull board

DAYJOB Health

  1. 48.3Claude Opus 5.5
  2. 34Claude Sonnet 5.5
  3. 31.2Claude Fable 5.1
  4. 31.2Muse Wonder
  5. 28.4GPT-6 Astra
  6. 27.6Grok 4.7
  7. 25.2GPT-6.1 Sol
  8. 23.6Gemini 4 Argon
  9. 22Claude Opus 5
  10. 22Muse Spark 1.3

15 modelsFull board

APEX-1 GP

  1. 73.04Claude Opus 5.5
  2. 72.12Claude Opus 5
  3. 68.36Gemini 3.8 Flash
  4. 64.35Kimi K3
  5. 63.75GPT-5.6 Sol
  6. 63.2GLM 5.3
  7. 61.03GPT-6 Astra
  8. 60.61GPT-6.1 Sol
  9. 59.38Grok 4.6
  10. 59.16Grok 4.7

12 modelsFull board

MLCR-AA

  1. 41.7Claude Sonnet 5.5
  2. 41.7GPT-6 Astra
  3. 38.3GPT-6.1 Sol
  4. 38.3Kimi K3.1 Pro
  5. 35.0Claude Haiku 5.5
  6. 33.9GPT-6 Sol
  7. 31.7GPT-5.5 Instant
  8. 30.6GPT-6.1 Terra
  9. 27.8Claude Opus 5.5
  10. 26.7DeepSeek V4.1 Flash 1017

27 modelsFull board

Bars are the model's lab colour. Each board's page says who ran every result.

Every model on every board

Index against price

48 models on 3 or more boards with a published list price. Log price scale. The dashed line joins models that no cheaper model beats.

020406080100$0.1$0.3$1$3$10$30$100Blended list price, USD per 1M tokensIndex scoreClaude Opus 5.5, Anthropic: 62.8 at $8.00 per 1M tokens, 8 of 12 boardsClaude Opus 5, Anthropic: 56.8 at $10.00 per 1M tokens, 10 of 12 boardsClaude Fable 5.1, Anthropic: 56.8 at $20.00 per 1M tokens, 6 of 12 boardsGPT-6 Astra, OpenAI: 56.1 at $20.00 per 1M tokens, 7 of 12 boardsGPT-6.1 Sol, OpenAI: 53.9 at $4.00 per 1M tokens, 7 of 12 boardsMuse Spark 1.1, Meta: 52.7 at $2.00 per 1M tokens, 3 of 12 boardsGemini 4 Argon, Google: 51.8 at $4.00 per 1M tokens, 6 of 12 boardsGPT-5.5, OpenAI: 51.7 at $11.25 per 1M tokens, 5 of 12 boardsGPT-5.6 Sol, OpenAI: 51.4 at $8.00 per 1M tokens, 9 of 12 boardsClaude Fable 5, Anthropic: 50.8 at $20.00 per 1M tokens, 6 of 12 boardsGPT-6 Luna, OpenAI: 50.6 at $0.20 per 1M tokens, 6 of 12 boardsClaude Opus 4.8, Anthropic: 50.6 at $10.00 per 1M tokens, 4 of 12 boardsKimi K3, Moonshot AI: 50.4 at $6.00 per 1M tokens, 7 of 12 boardsGrok 4.7, SpaceX AI: 50.0 at $3.00 per 1M tokens, 7 of 12 boardsClaude Sonnet 5, Anthropic: 49.8 at $4.00 per 1M tokens, 6 of 12 boardsGPT-6 Sol, OpenAI: 49.6 at $4.00 per 1M tokens, 7 of 12 boardsGemini 3.6 Flash, Google: 49.1 at $1.50 per 1M tokens, 3 of 12 boardsGPT-5.6 Terra, OpenAI: 48.9 at $4.50 per 1M tokens, 4 of 12 boardsClaude Opus 4.6, Anthropic: 48.0 at $10.00 per 1M tokens, 5 of 12 boardsClaude Opus 4.5, Anthropic: 47.7 at $10.00 per 1M tokens, 3 of 12 boardsGLM 5.3, Zhipu: 47.4 at $2.15 per 1M tokens, 7 of 12 boardsGemini 3.8 Flash, Google: 47.3 at $1.50 per 1M tokens, 7 of 12 boardsClaude Opus 4.7, Anthropic: 47.3 at $10.00 per 1M tokens, 4 of 12 boardsGPT-5.6 Luna, OpenAI: 47.3 at $0.45 per 1M tokens, 4 of 12 boardsQwen3.8 Max, Alibaba: 47.0 at $2.48 per 1M tokens, 4 of 12 boardsGPT-5, OpenAI: 46.1 at $3.44 per 1M tokens, 4 of 12 boardsGPT-5 mini, OpenAI: 45.9 at $0.69 per 1M tokens, 3 of 12 boardsGemini 2.5 Pro, Google: 45.7 at $3.44 per 1M tokens, 3 of 12 boardsGrok 4.6, SpaceX AI: 45.5 at $3.00 per 1M tokens, 6 of 12 boardsGPT-5.2, OpenAI: 45.5 at $4.81 per 1M tokens, 4 of 12 boardsDeepSeek V4.1 Flash, DeepSeek: 45.3 at $0.52 per 1M tokens, 6 of 12 boardsMuse Spark 1.2, Meta: 44.9 at $2.00 per 1M tokens, 3 of 12 boardsGPT-5.4, OpenAI: 44.5 at $5.63 per 1M tokens, 5 of 12 boardsGrok 4.3, SpaceX AI: 43.5 at $1.56 per 1M tokens, 3 of 12 boardsGPT-5.1, OpenAI: 43.2 at $3.44 per 1M tokens, 3 of 12 boardsGemini 3.1 Pro, Google: 42.4 at $4.50 per 1M tokens, 6 of 12 boardsHy4 Preview, Tencent: 42.3 at $1.25 per 1M tokens, 3 of 12 boardsGLM 5.1, Zhipu: 42.2 at $2.15 per 1M tokens, 3 of 12 boardsDeepSeek V4 Pro, DeepSeek: 41.8 at $1.98 per 1M tokens, 5 of 12 boardsClaude Sonnet 4.6, Anthropic: 41.5 at $6.00 per 1M tokens, 3 of 12 boardsInkling, Thinking Machines: 40.6 at $1.76 per 1M tokens, 3 of 12 boardsKimi K2.6, Moonshot AI: 39.9 at $1.71 per 1M tokens, 4 of 12 boardsQwen3.6 Plus, Alibaba: 37.9 at $0.62 per 1M tokens, 4 of 12 boardsMiniMax-M2.7, MiniMax: 35.3 at $0.52 per 1M tokens, 3 of 12 boardsGrok 4.20, SpaceX AI: 32.6 at $1.56 per 1M tokens, 4 of 12 boardsClaude Haiku 4.5, Anthropic: 31.0 at $2.00 per 1M tokens, 5 of 12 boardsClaude Haiku 5.5, Anthropic: 55.5 at $0.20 per 1M tokens, 4 of 12 boardsClaude Haiku 5.5 55.5Claude Sonnet 5.5, Anthropic: 63.2 at $4.00 per 1M tokens, 7 of 12 boardsClaude Sonnet 5.5 63.2Price frontier

How the index is made

Each model's page shows its full calculation.

  1. One scale. Each result is read on its own board's 0 to 100 scale.
  2. Easy and hard boards. Each board's scores are read against how the same models do elsewhere, so a model is never marked down for a board it has no result on.
  3. Weights. Boards that measure clinical work most directly count most; boards near their ceiling count less.
  4. Coverage. A model with few boards is pulled toward the field average of 44.2, so the number of boards appears next to every score.

Read the method

Each index score, with the board results beneath it

A forest plot of the Clinical Benchmarks Index. Each row is one model. Squares are its results on the index boards, each on that board's own 0 to 100 scale; the diamond is its index score. Ranks are across all 128 models.

Index boards behind the score25 of 128 ranked models shown
74 models with fewer than 3 boards are hidden by this filter; their ranks still count.
Index score Board result, run by the publisher or an independent group Board result reported by the model's own lab Spread of its results Field average, 44.2
  1. 1Claude Sonnet 5.5Anthropic · 7 of 12 boards63.27/12
  2. 2Claude Opus 5.5Anthropic · 8 of 12 boards62.88/12
  3. 3Claude Opus 5Anthropic · 10 of 12 boards56.810/12
  4. 4Claude Fable 5.1Anthropic · 6 of 12 boards56.86/12
  5. 5GPT-6 AstraOpenAI · 7 of 12 boards56.17/12
  6. 6Claude Haiku 5.5Anthropic · 4 of 12 boards55.54/12
  7. 7GPT-6.1 SolOpenAI · 7 of 12 boards53.97/12
  8. 8Muse Spark 1.1Meta · 3 of 12 boards52.73/12
  9. 10Gemini 4 ArgonGoogle · 6 of 12 boards51.86/12
  10. 11GPT-5.5OpenAI · 5 of 12 boards51.75/12
  11. 13Muse WonderMeta · 3 of 12 boards51.53/12
  12. 14GPT-5.6 SolOpenAI · 9 of 12 boards51.49/12
  13. 15Claude Fable 5Anthropic · 6 of 12 boards50.86/12
  14. 16GPT-6 LunaOpenAI · 6 of 12 boards50.66/12
  15. 17Claude Opus 4.8Anthropic · 4 of 12 boards50.64/12
  16. 18Kimi K3Moonshot AI · 7 of 12 boards50.47/12
  17. 20Grok 4.7SpaceX AI · 7 of 12 boards50.07/12
  18. 21Claude Sonnet 5Anthropic · 6 of 12 boards49.86/12
  19. 23GPT-6 SolOpenAI · 7 of 12 boards49.67/12
  20. 24Muse SparkMeta · 4 of 12 boards49.64/12
  21. 26Gemini 3.6 FlashGoogle · 3 of 12 boards49.13/12
  22. 27GPT-5.6 TerraOpenAI · 4 of 12 boards48.94/12
  23. 32Claude Opus 4.6Anthropic · 5 of 12 boards48.05/12
  24. 34Claude Opus 4.5Anthropic · 3 of 12 boards47.73/12
  25. 35GLM 5.3Zhipu · 7 of 12 boards47.47/12

Method of 31 October 2026. Select a square to read that result's quote and page.

How it is computed Index as CSV

Every board, on its own scale

The top 15 results on each board, in that board's rank order and on its own scale. The panels do not share an axis because the boards measure different things.

Clinical reasoning and knowledgeMAST (Medical AI Superintelligence Test)A composite of clinical benchmarks spanning diagnostic and management reasoning, safety, multimodal imaging, and agentic capability.1000GPT-5.6 Sol: 60.2Kimi K3: 60.1Gemini 3.6 Flash: 59.3Gemini 3.1 Pro: 58.9Qwen3.5 397B A17B: 57.9Claude Opus 5: 57.1Claude Sonnet 5: 56.6Grok 4.3: 53.78 results · to Aug 2026GPT-5.6 Sol 60.2
Clinical reasoning and knowledgeAPEX-1 GPPrimary-care consults graded by practising GPs on diagnosis, workup and safe escalation.1000Claude Opus 5.5: 73.04Claude Opus 5: 72.12Gemini 3.8 Flash: 68.36Kimi K3: 64.35GPT-5.6 Sol: 63.75GLM 5.3: 63.2GPT-6 Astra: 61.03GPT-6.1 Sol: 60.61Grok 4.6: 59.38Grok 4.7: 59.16DeepSeek V4 Pro 0813: 56.53DeepSeek V4.1 Flash: 56.1812 results · to Oct 2026Claude Opus 5.5 73.04

Bars are the model's lab colour. Each board's page says who ran every result.

All boards with details

Where the numbers come from

Each count is a count of rows in the published dataset.

92

source documents, each held with a verbatim quote and a page locator

58official leaderboard
16system card
6model card
5launch post
7other kinds
472

results read from those documents

361published by the board's own maintainers
49run by an independent group
62reported by the model's own lab
14

clinical boards tracked

12boards in the index
Left out of the index
  • HealthAgentBench: fewer than 5 models
  • CHI-Bench: fewer than 5 models
128

models ranked by the index

54rest on 3 or more boards
46rest on 2 boards
28rest on a single board

A result is added only with its source document, a verbatim quote and a locator. Results reported by the model's own lab are included and drawn hollow throughout.

Every source document