HealthBench Hard Leaderboard

Last updated September 8, 2026 · 16 models evaluated

Muse Spark leads the HealthBench Hard leaderboard at 0.428, ahead of GPT-6 Astra (0.363) and GPT-5 (0.347). HealthBench Hard is the hardest slice of OpenAI's HealthBench benchmark: the 1,000 health conversations frontier models handled worst, graded against physician-written rubrics on a 0 to 1 scale. The best score at the benchmark's May 2025 release was o3's 0.320.

1,000-conversation set, scores as published by each document. Updated September 8, 2026.

#	model		score	size	context	cost in / out per 1M	license
1		Muse Spark Meta via Introducing Muse Spark: Scaling Towards Personal Superintelligence	0.428	—	1.0M	$1.25 / $4.25	proprietary
2		GPT-6 Astra OpenAI via GPT-6 Astra System Card	0.363	—	1.05M	$10.00 / $50.00	proprietary
3		GPT-5 OpenAI via GPT-5.6 System Card	0.347	—	400K	$1.25 / $10.00	proprietary
4		GPT-5.2 OpenAI via GPT-5.6 System Card	0.343	—	400K	$1.75 / $14.00	proprietary
5		GPT-5.6 Sol OpenAI via GPT-5.6 System Card	0.331	—	1.1M	$5.00 / $30.00	proprietary
6		GPT-5.6 Terra OpenAI via GPT-5.6 System Card	0.327	—	1.1M	$2.00 / $12.00	proprietary
7		GPT-5.6 Luna OpenAI via GPT-5.6 System Card	0.320	—	1.1M	$0.20 / $1.20	proprietary
8		GPT-5.5 OpenAI via GPT-5.6 System Card	0.315	—	1.05M	$5.00 / $30.00	proprietary
9		GPT-5.6 Sol (August) OpenAI via GPT-5.6 - August Updates (system card addendum)	0.314	—	1.1M	$5.00 / $30.00	proprietary
10		GPT OSS 120B OpenAI via gpt-oss-120b & gpt-oss-20b Model Card	0.300	117B	131K	—	open
11		GPT-5.4 OpenAI via GPT-5.6 System Card	0.291	—	1.05M	$2.50 / $15.00	proprietary
12		GPT-5.6 Luna (August) OpenAI via GPT-5.6 - August Updates (system card addendum)	0.287	—	1.1M	$0.20 / $1.20	proprietary
13		GPT-5.3 Chat OpenAI via GPT-5.3 Instant System Card	0.259	—	128K	$1.75 / $14.00	proprietary
14		GPT-5.1 OpenAI via GPT-5.6 System Card	0.254	—	400K	$1.25 / $10.00	proprietary
15		GPT-5.5 Instant OpenAI via GPT-5.5 Instant System Card	0.229	—	400K	$5.00 / $30.00	proprietary
16		GPT OSS 20B OpenAI via gpt-oss-120b & gpt-oss-20b Model Card	0.108	21B	131K	—	open

Scores are rubric credit across the 1,000-conversation set, copied exactly as the document under each model name prints them. The line under a name says which document, and links to its entry on the sources page; "source pending" marks a row whose document has not been located or verified yet. Prices are vendor list rates per 1M tokens, checked September 8, 2026. GPT-5.6 models charge higher rates above 272K input tokens. GPT OSS models are open weights without vendor list pricing. Every model name links through to its full result.

HealthBench Hard keeps only the failures. OpenAI's HealthBench grades models on 5,000 realistic health conversations, from emergency triage to global health, against 48,562 rubric criteria written by 262 physicians across 60 countries. The Hard subset is the 1,000 conversations on which frontier models scored worst at release, which makes it the part of HealthBench that still separates current models. Two open-weights models, GPT OSS 120B and GPT OSS 20B, sit on the board alongside the closed frontier. The full construction is on the benchmark page.

A score is rubric points earned over maximum positive points, clipped to 0 to 1. Low numbers are the design, not a flaw: the subset exists because models fail on it, the field's best result is 0.428, and 1.0 is nowhere in sight. The numbers here are compiled, not measured by this site: each row is read from a published document, most often the vendor's own system card, and the sources page shows the quote it came from. Two documents can grade the same model differently, for reasons the methodology page lays out, so a gap between two rows read from different documents carries that caveat.

The pairings worth deciding between get full pages; the compare page holds the complete score-difference matrix.

• Muse Spark vs GPT-5.6 Sol
  0.428 vs 0.331 · Muse Spark by 0.097
• GPT-5.6 Sol vs GPT-5.6 Terra
  0.331 vs 0.327 · GPT-5.6 Sol by 0.004
• GPT-5.6 Luna vs GPT OSS 120B
  0.320 vs 0.300 · GPT-5.6 Luna by 0.020
• GPT OSS 120B vs GPT OSS 20B
  0.300 vs 0.108 · GPT OSS 120B by 0.192
• GPT-5 vs GPT-5.6 Sol
  0.347 vs 0.331 · GPT-5 by 0.016
• GPT-5.6 Luna vs GPT-5.3 Chat
  0.320 vs 0.259 · GPT-5.6 Luna by 0.061

Everything on this page ships as JSON and CSV under CC BY 4.0, at URLs that never move. The data page has the citation format and version history.