HealthBench
Can the model produce useful work in a professional knowledge-work setting?
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 58.3% | Reported configurationOpenAI report1 value · 1 reportSep 3, 2026 · source |
| 2 | GPT-5.6 SolOpenAI | 57.0% | Reported configurationOpenAI report1 value · 1 reportSep 3, 2026 · source |
| 3 | GPT-5.6 TerraOpenAI | 57.0% | Reported configurationOpenAI report1 value · 1 reportSep 3, 2026 · source |
| 4 | GPT-5.5OpenAI | 56.5% | Reported configurationOpenAI report1 value · 1 reportSep 3, 2026 · source |
| 5 | GPT-5.6 LunaOpenAI | 55.8% | Reported configurationOpenAI report1 value · 1 reportSep 3, 2026 · source |
| 6 | GPT-6 LunaOpenAI | 54.5% | Reported configurationOpenAI report1 value · 1 reportSep 22, 2026 · source |
| 7 | GPT-6 SolOpenAI | 53.2% | Reported configurationOpenAI report1 value · 1 reportSep 22, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
Definition and comparison boundarypublic methodology
Can the model produce useful work in a professional knowledge-work setting?
Professional analysis, document, finance, office, or domain-specific tasks. Higher is better. Length-adjusted score on the 0-100 scale reported by OpenAI.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
OpenAI's September 22 appendix corrects the earlier Astra value to 58.3. The table prints length-adjusted score followed by unadjusted score and mean response length in characters: GPT-6 Astra 58.3 (60.8, 977); GPT-6 Sol 53.2 (47.1, 977); GPT-6 Luna 54.5 (50.0, 1255). Earlier GPT-5.5 and GPT-5.6 comparison values are retained from the same report: 56.5 (58.4, 2313); 57.0 (55.6, 1764); 57.0 (58.7, 2285); 55.8 (55.4, 1930).
Method / source