Skip to main content
BenchLM
Data
Specialist vertical · BenchAlign v6.0 · 36 ranked (0 Supported, 36 Estimated)
Scored from the dataset of

Healthcare and medical leaderboard

As of November 21, 2026, Claude Fable 5.2 leads our healthcare and medical leaderboard at 87.5 (Estimated, $8/$40 per 1M tokens), ahead of Gemini 4 Carbon at 85.7. All 36 rows currently carry Estimated labels; Gemini 4 Carbon (#2, 85.7) is the lowest-cost model near the top at $2/$10 per 1M tokens; DeepSeek-R1 leads open-weight models at #32 (43.8). We rank 36 models across 17 domain protocols, and this vertical carries zero weight in the overall score.

Shortlist by workload and budget

Each pick comes directly from the live healthcare & medicine rows below, with listed API rates and the case where another option wins.

Highest point score#1 · Estimated

Anthropic · $8/$40 per 1M tokens

Carries an Estimated label because independent multi-evaluator coverage on healthcare & medicine is still thin.

Closest contender#2 · Estimated

Google · $2/$10 per 1M tokens

Sits 1.8 points behind Claude Fable 5.2; test both on your own prompts when provider lock-in or latency matters.

Best API value#2 · Estimated

Google · $2/$10 per 1M tokens

Keeps 98% of the top score (85.7 vs 87.5), but gives up 1.8 points on the hardest tail cases.

Top open-weight model#32 · Estimated

DeepSeek · Open weights · $0.55/$2.19 per 1M tokens

Trails Claude Fable 5.2 by 43.7 points (#32 overall here), so pick it when weight control or on-prem residency outweighs peak score.

Ranked models

Ordered by BenchAlign v6.0 point estimate as of November 21, 2026. An Estimated row backed by fewer than two source families never ranks above a row scoring within 2 points below it. Conditional ranges do not establish rank confidence.

Healthcare & Medicine LLM Leaderboard
RankModelScoreEvidence
#1Anthropic87.5Estimated
#2Google85.7Estimated
#3Anthropic82.7Estimated
#4OpenAI83.9Estimated
#5Anthropic79.2Estimated
#6OpenAI77.1Estimated
#7OpenAI77.7Estimated
#8OpenAI75.8Estimated
#9Anthropic74.0Estimated
#10Meta72.6Estimated
#11OpenAI71.7Estimated
#12Anthropic71.0Estimated
#13Alibaba68.4Estimated
#14OpenAI66.9Estimated
#15OpenAI66.6Estimated
#16Meta63.7Estimated
#17OpenAI61.2Estimated
#18Google59.8Estimated
#19OpenAI61.4Estimated
#20OpenAI61.0Estimated
#21Meta58.0Estimated
#22OpenAI55.5Estimated
#23OpenAI56.0Estimated
#24Meta52.5Estimated
#25Anthropic52.0Estimated
#26Anthropic52.0Estimated
#27OpenAI48.4Estimated
#28Alibaba49.8Estimated
#29OpenAI46.4Estimated
#30xAI45.5Estimated
#31Google44.2Estimated
#32
DeepSeek-R1Open weights
DeepSeek
43.8Estimated
#33Anthropic41.6Estimated
#34OpenAI34.6Estimated
#35OpenAI27.5Estimated
#36Google24.7Estimated

Scored domain benchmarks (17)

Each percentage is the protocol's relative reference weight inside this vertical's calibrated BenchAlign v6.0 model, not a raw score average. Independent evaluator shares are capped at 40% whenever three or more independent evaluators contribute.

An open benchmark for clinician-facing model responses across care consult, writing and documentation, and medical research tasks.

19% ref. weight

A harder subset of OpenAI's HealthBench for evaluating open-ended medical and health reasoning with rubric-based grading.

10% ref. weight

Raw score on physician-authored clinical consult, documentation, and research conversations.

9% ref. weight

HealthBench score after applying a verbosity penalty to model responses.

8% ref. weight

Raw score on realistic multi-turn healthcare conversations graded against expert-written rubrics.

8% ref. weight

A medical multiple-choice benchmark spanning many specialties with 10 answer options per question.

7% ref. weight

A public benchmark of 100 physician tasks carried out inside an electronic health record system.

7% ref. weight

Valen Labs' benchmark of AI agents doing machine learning work in computational biology: 25 tasks in antibody development, designed binders, protein engineering, structural biology and functional genomics. Each agent builds a model from experimental data in an offline sandbox, and its prediction script is scored on held-out measurements, from 0 (a simple baseline) to 100 (a perfect answer), over three runs per task.

6% ref. weight

A multimodal medical multiple-choice benchmark covering clinical images such as X-rays, histology, and dermatology.

4% ref. weight
MedHELM: Administration & Workflow· Stanford CRFM MedHELM

Scheduling, referral routing, prior-authorization, and hospital administrative workflow tasks.

3% ref. weight
MedHELM: Clinical Decision Support· Stanford CRFM MedHELM

Evidence-based differential diagnosis, triage, and treatment-plan support tasks.

3% ref. weight
MedHELM: Clinical Note Generation· Stanford CRFM MedHELM

Structured SOAP note drafting and visit summarization from clinical encounters.

3% ref. weight
MedHELM: MedCalc-Bench· Stanford CRFM MedHELM

Clinical equation selection and bedside medical score computation from patient charts.

3% ref. weight
MedHELM: MedHallu· Stanford CRFM MedHELM

Detection and avoidance of fabricated medical claims in clinical question answering.

3% ref. weight
Stanford MedHELM (overall win rate)· Stanford CRFM MedHELM

Aggregate head-to-head win rate across Stanford CRFM MedHELM clinical task categories.

3% ref. weight
MedHELM: PubMedQA· Stanford CRFM MedHELM

Biomedical research question answering grounded in PubMed study abstracts.

3% ref. weight
MedHELM: MedQA· Stanford CRFM MedHELM

USMLE-style medical licensing examination questions under the MedHELM harness.

1% ref. weight

Physician-graded clinical rubrics paired with Stanford MedHELM hospital tasks

Seventeen protocols feed the healthcare vertical, led by HealthBench Professional (19%), HealthBench Hard (10%), HealthBench Professional (raw) (9%) alongside MedXpertQA (text and multimodal), PhysicianBench, FrontierBioML, and eight Apache-2.0 Stanford CRFM MedHELM tracks covering clinical decision support, SOAP note generation, MedCalc-Bench bedside equations, MedHallu hallucination resistance, PubMedQA, and administrative workflows.

Combining multi-turn physician conversation rubrics with hospital paperwork and equation calculators prevents models memorized on multiple-choice USMLE flashcards from dominating practical clinical-documentation rankings.

Estimated evidence status, clinical safety limits, and HIPAA contracting

All 36 models on this board currently carry Estimated labels (0 Supported) because independent clinical evaluators test smaller, only partly overlapping model subsets. Treat small point differences cautiously until broader multi-evaluator overlap lands.

No benchmark score on this page establishes clinical safety, diagnostic reliability, FDA software-as-a-medical-device clearance, or HIPAA eligibility. Before sending protected health information to any API ranked here, verify Business Associate Agreement terms, zero-data-retention endpoints, and clinician signoff workflows on our HIPAA provider ledger.

Weighing clinical rubric accuracy against API cost and PHI residency

Claude Fable 5.2 (87.5, $8/$40 per 1M tokens) leads Gemini 4 Carbon (85.7, $2/$10 per 1M tokens) by 1.8 points, with Claude Opus 5.5 third at 82.7. For high-volume ambient scribing, referral triage, or prior-authorization drafting under a signed BAA, Gemini 4 Carbon delivers 98% of the leader's score at one-quarter of the per-token price.

Health systems restricted to on-premises weights face a steeper drop-off here: DeepSeek-R1 is the highest ranked open-weight row at #32 (43.8) primarily because fewer recent open releases have been run through HealthBench and MedHELM, so validate newer open checkpoints directly on your own de-identified notes.

Questions

Which LLM is best for healthcare and medicine right now?

Claude Fable 5.2 holds #1 on our healthcare and medicine leaderboard at 87.5 (Estimated) as of November 21, 2026, ahead of Gemini 4 Carbon at 85.7 and Claude Opus 5.5 at 82.7. Every ranked row currently carries an Estimated label, so validate top contenders on your own workload.

What is the best open-weight LLM for healthcare and medicine?

DeepSeek-R1 from DeepSeek leads open-weight models on this healthcare and medicine board at #32 with a calibrated score of 43.8 (Estimated). That trails Claude Fable 5.2 by 43.7 points, so choose DeepSeek-R1 when self-hosting, VPC data residency, or weight customization matters more than peak frontier accuracy.

Which API model offers the best value for healthcare and medicine?

Gemini 4 Carbon (#2, 85.7) is our top API value pick for healthcare and medicine, priced at $2/$10 per 1M tokens compared with Claude Fable 5.2 at $8/$40 per 1M tokens. You give up 1.8 calibrated points, making Gemini 4 Carbon the better buy whenever those extra points do not cut downstream retries.

Which benchmarks make up the healthcare and medicine leaderboard score?

BenchAlign v6.0 calibrates 17 open and primary domain protocols for this board, led by HealthBench Professional (19%), HealthBench Hard (10%), HealthBench Professional (raw) (9%), HealthBench (length-adjusted) (8%). Each percentage reflects share of the vertical's reference weight inside our item-response model rather than a raw score average, and no single independent evaluator can exceed 40% when three or more evaluators contribute.

Does a high healthcare benchmark score mean an LLM is HIPAA compliant?

No. HealthBench, PhysicianBench, and Stanford MedHELM test clinical reasoning and documentation quality, whereas HIPAA compliance depends on legal Business Associate Agreements, encryption, and zero-data-retention API endpoints. Check our /hipaa-compliant-ai provider directory to verify which vendor tiers sign BAAs before transmitting protected health information.

Why are all models on the healthcare leaderboard labeled Estimated?

All 36 healthcare rows currently carry Estimated labels because public clinical evaluations—spanning OpenAI HealthBench rubrics, MedXpertQA, PhysicianBench, and Stanford CRFM MedHELM—test partially overlapping model cohorts. Scores still separate strong clinical communicators from weaker ones, but small point gaps do not establish clinical superiority.

Healthcare & medicine leaderboard updates

Weekly updates when new models or domain evaluations move the healthcare & medicine rankings.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.