# HealthBench Hard leaderboard

Knowledge benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/knowledge-health-bench-hard

Every model the catalog carries a published HealthBench Hard value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Knowledge |
| Measure | Open-ended health evaluation |
| Tasks | 1,000 health prompts |
| Difficulty | Advanced health reasoning |
| Published by | [Muse Spark Eval Methodology](https://ai.meta.com/static-resource/muse-spark-eval-methodology) |

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

## HealthBench Hard ranking

13 models with a published HealthBench Hard value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Open-ended health evaluation | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Muse Spark](https://modelscale.dev/models/muse-spark) | Meta | 42.8 | Observed 2026-11-21 · source benchlm:models |
| 2 | [GPT-5.4](https://modelscale.dev/models/gpt-5-4) | OpenAI | 40.1 | Observed 2026-11-21 · source benchlm:models |
| 3 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 36.6 | Observed 2026-11-21 · source benchlm:models |
| 4 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 36.2 | Observed 2026-11-21 · source benchlm:models |
| 5 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 33.1 | Observed 2026-11-21 · source benchlm:models |
| 6 | [GPT-5.6 Terra](https://modelscale.dev/models/gpt-5-6-terra) | OpenAI | 32.7 | Observed 2026-11-21 · source benchlm:models |
| 7 | [GPT-5.6 Luna](https://modelscale.dev/models/gpt-5-6-luna) | OpenAI | 32.0 | Observed 2026-11-21 · source benchlm:models |
| 8 | [GPT-6 Luna](https://modelscale.dev/models/gpt-6-luna) | OpenAI | 31.4 | Observed 2026-11-21 · source benchlm:models |
| 8 | [GPT-6.1 Terra](https://modelscale.dev/models/gpt-6-1-terra) | OpenAI | 31.4 | Observed 2026-11-21 · source benchlm:models |
| 10 | [GPT-6 Sol](https://modelscale.dev/models/gpt-6-sol) | OpenAI | 30.1 | Observed 2026-11-21 · source benchlm:models |
| 11 | [Gemini 3.1 Pro](https://modelscale.dev/models/gemini-3-1-pro) | Google | 20.6 | Observed 2026-11-21 · source benchlm:models |
| 12 | [Grok 4.20](https://modelscale.dev/models/grok-4-20-beta) | xAI | 20.3 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Claude Opus 4.6](https://modelscale.dev/models/claude-opus-4-6) | Anthropic | 14.8 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
