Medical Evals

We benchmarked today's leading AI models on MedXpertQA, a dataset of expert-level medical exam questions spanning diagnosis, treatment, and basic science — including a multimodal subset with real clinical images. Every model ran zero-shot with chain-of-thought reasoning, no fine-tuning or few-shot examples.

Google GeminiGoogle
OpenAIOpenAI
AnthropicAnthropic

Multimodal medical reasoning

2,000 questions with clinical images · 5 answer choices · Higher is better

0%25%50%75%100%Gemini 3.8 Flash: 86.60% (1,731/2,000)86.6%Gemini 3.8 FlashGemini 3.1 Pro Preview: 82.00% (1,640/2,000)82.0%Gemini 3.1 Pro Previewgpt-5.6-sol: 76.65% (1,533/2,000)76.7%gpt-5.6-solClaude Opus 5: 76.00% (1,521/2,000)76.0%Claude Opus 5Claude Sonnet 5: 58.65% (1,173/2,000)58.6%Claude Sonnet 5GPT-4o: 42.80%42.8%GPT-4o

Text-only medical reasoning

2,450 questions · 10 answer choices · Higher is better

0%25%50%75%100%Gemini 3.8 Flash: 66.20% (1,621/2,450)66.2%Gemini 3.8 Flashgpt-5.6-sol: 59.55% (1,459/2,450)59.5%gpt-5.6-solClaude Sonnet 5: 49.35% (1,209/2,450)49.4%Claude Sonnet 5GPT-4o: 30.37%30.4%GPT-4o

Not every model was run on both subsets. Scores are single runs at each provider's default settings (no reasoning-effort tuning), graded by exact-match on the model's final answer letter. GPT-4o's score is from the original MedXpertQA paper's published leaderboard rather than a run in this project — it's directly comparable because it used the same evaluation harness as every other score here.

Where models struggle

Breaking multimodal accuracy down by organ system and by task type shows the gap between models isn't uniform — some organ systems are hard for every model, while the ranking between models holds fairly steady across task types.

Accuracy by organ system

Multimodal subset · darker = higher accuracy

LymphaticNervousUrinaryMuscularReproductiveIntegumentaryEndocrineSkeletalDigestiveOther / NACardiovascularRespiratory
Gemini 3.8 Flash949291918988888786838181
Gemini 3.1 Pro Preview908482898583828280677781
gpt-5.6-sol807470877780777971677478
Claude Opus 5897780807480787574617375
Claude Sonnet 5695763656170625655445751
44%
94%accuracy

Accuracy by task type

Multimodal subset · Higher is better

Gemini 3.8 Flash
Gemini 3.1 Pro Preview
gpt-5.6-sol
Claude Opus 5
Claude Sonnet 5
0255075100Gemini 3.8 Flash — Diagnosis: 86.4%86Gemini 3.1 Pro Preview — Diagnosis: 81.3%81gpt-5.6-sol — Diagnosis: 75.9%76Claude Opus 5 — Diagnosis: 74.6%75Claude Sonnet 5 — Diagnosis: 57.8%58DiagnosisGemini 3.8 Flash — Treatment: 85.5%85Gemini 3.1 Pro Preview — Treatment: 82.6%83gpt-5.6-sol — Treatment: 81.9%82Claude Opus 5 — Treatment: 77.7%78Claude Sonnet 5 — Treatment: 62.7%63TreatmentGemini 3.8 Flash — Basic Science: 88.4%88Gemini 3.1 Pro Preview — Basic Science: 83.6%84gpt-5.6-sol — Basic Science: 72.5%73Claude Opus 5 — Basic Science: 79.0%79Claude Sonnet 5 — Basic Science: 56.4%56Basic Science

Limitations

  • The dataset itself has some construction issues. MedXpertQA expands each question's original answer choices up to 10 using an LLM-assisted process, which occasionally produces two choices describing the same underlying fact, a handful of internally contradictory questions, and — in the multimodal subset — the occasional low-resolution or mismatched image. In manual review, roughly 17–21% of the answers models got “wrong” had a real, defensible argument for the model's answer. True accuracy for every model here is likely a few points higher than the raw numbers shown.
  • Single run, default settings. Each score is one pass per model at the provider's default reasoning effort, not an average across repeated trials or a best-of-N result.
  • Organ-system and task-type breakdowns use smaller samples. Some organ systems have as few as 18 questions, so those individual cells carry more statistical noise than the headline scores above.
  • Grading is exact-match on the model's stated letter. A correctly-reasoned answer expressed in an unexpected format can, in rare cases, be mis-scored by the automated grader.