HealthBench
AI benchmark results and model performance
Type: percentageHealth
Recorded results
23
Average score
54.33%
Score range
30% - 67.20%
Leading model
67.20% - GPT 5
Scores over time
Individual benchmark scores plotted by date.
Model results
| Organisation | Models | Reported | Top score | Info | Self reported | Source |
|---|---|---|---|---|---|---|
| GPT 5 | 07 Aug 2025 | 67.20% | With Thinking | Yes | Source | |
| Claude Opus 5 | 24 Jul 2026 | 67.10% | Raw score; adaptive thinking at max effort; no tools or custom system prompt; Claude Opus 4.8 grader; average of 5 trials | Yes | Source | |
| Claude Fable 5 | 09 Jun 2026 | 66% | Professional; starred row in Anthropic's launch table | Yes | Source | |
| Claude Mythos 5 | 09 Jun 2026 | 66% | Professional; starred row in Anthropic's launch table | Yes | Source | |
| GPT-6.1 Sol | 29 Sept 2026 | 58.50% | HealthBench; length-adjusted score (unadjusted 56.7, mean response length 1701 chars); GPT-6.1 Terra system card | Yes | Source | |
| GPT-6 Astra | 03 Sept 2026 | 58.30% | HealthBench; length-adjusted score (unadjusted 56.9, mean response length 1760 chars); GPT-6.1 Terra system card | Yes | Source | |
| Kimi K2 Thinking | 06 Nov 2025 | 58% | inferred alias from kimi-k2-thinking-0905 | Yes | Source | |
| Seed 2.0 Pro | 14 Feb 2026 | 57.70% | Seed2 official benchmark table | HealthBench | Yes | Source | |
| GPT OSS 120b | 05 Aug 2025 | 57.60% | High Reasoning Effort | Yes | Source | |
| GPT OSS Safeguard 120b | 29 Oct 2025 | 57.60% | inferred high-confidence family alias from gpt-oss-120b (score=0.5102; benches=7) | Yes | Source | |
| GPT 5.6 Terra | 09 Jul 2026 | 57% | HealthBench; length-adjusted score (unadjusted 58.7, mean response length 2285 chars); GPT-6.1 Terra system card | Yes | Source | |
| GPT-6.1 Terra | 26 Oct 2026 | 55.10% | HealthBench; length-adjusted score (unadjusted 51.1, mean response length 1331 chars); GPT-6.1 Terra system card | Yes | Source | |
| GPT 5.3 Chat | 03 Mar 2026 | 54.10% | - | Yes | Source | |
| Seed 2.0 Lite | 14 Feb 2026 | 51.20% | Seed2 official benchmark table | HealthBench | Yes | Source | |
| GPT OSS 20b | 05 Aug 2025 | 42.50% | High Reasoning Effort | Yes | Source | |
| GPT OSS Safeguard 20b | 29 Oct 2025 | 42.50% | inferred high-confidence family alias from gpt-oss-20b (score=0.5137; benches=7) | Yes | Source | |
| Seed 2.0 Mini | 14 Feb 2026 | 30% | Seed2 official benchmark table | HealthBench | Yes | Source |
