Benchmark
HealthBench Hard
What is HealthBench Hard?
A challenging variation of HealthBench that evaluates large language models' performance and safety in healthcare through 5,000 multi-turn conversations with…
- Released
- 2025-05-12
- Modality
- text
- Evaluates
- healthcare
- Openness
- unknown
- Importer
- LLM Stats
- Reported scores
- 9
Source provenance
HealthBench Hard results and reported scores
LLM Stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| GPT OSS 120B | OpenAI | 0.3 | 2025-08-05 | evidence |
| GPT OSS 20B | OpenAI | 0.108 | 2025-08-05 | evidence |
| GPT-5 | OpenAI | 0.016 | 2025-08-07 | evidence |
| GPT-5.3 Chat | OpenAI | 0.259 | 2026-03-04 | evidence |
| GPT-5.5 Instant | OpenAI | 0.229 | 2026-05-05 | evidence |
| GPT-5.6 Luna | OpenAI | 0.32 | 2026-07-09 | evidence |
| GPT-5.6 Sol | OpenAI | 0.331 | 2026-07-09 | evidence |
| GPT-5.6 Terra | OpenAI | 0.327 | 2026-07-09 | evidence |
| Muse Spark | Meta | 0.428 | 2026-04-08 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.