HealthBench Professional (length-adjusted score)
OpenAI medical conversation evaluation (for experts). Length-adjusted score.
- Category
- Knowledge
- Unit
- Score
- Higher is better
- Comparison groups
- 3total
- Currently available model rankings · Ranked results: 8
- All models (records): Groups: 3 · Entries: 11
Rankings
Values with different source types (official, third-party, paper) or measurement conditions cannot be compared, so rankings are separated by comparison group (the same source, or sources verified to use the same conditions).
Rankings include only models currently available in Claude Code / Codex, excluding legacy models.
official results (GPT-6 Astra System Card (Table 6. HealthBench Evaluations))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra | not stated | 64.7 | — | OpenAI | GPT-6 Astra System Card (Table 6. HealthBench Evaluations)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoreSource excerpt64.7Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row |
official results (Claude Fable 5.2 and Claude Mythos 5.2 System Card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Max | 67.7 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt67.7%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 2 | Claude Sonnet 5.5 | Max | 63 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt63.0%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 3 | Claude Opus 5.5 | Max | 59.9 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt59.9%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 4 | Claude Haiku 5.5 | Max | 58.4 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback status is not stated for this model. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt58.4%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
official results (Addendum to GPT-6 Astra System Card: GPT-6.1 Terra)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | GPT-6.1 Sol | not stated | 64.2 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt64.2 (67.2, 3038)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-1-sol column |
| 2 | GPT-6.1 Terra | not stated | 62.7 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt62.7 (64.2, 2531)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-1-terra column |
| 3 | GPT-6 Luna | not stated | 60.8 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt60.8 (61.2, 2119)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-luna column |
All model rankings (records)
These rankings are recalculated by comparison group including legacy models and models unavailable in Claude Code / Codex. They differ from the rankings above because the set of models differs.
Show rankings including legacy and unavailable models (entries: 11)
official results (GPT-6 Astra System Card (Table 6. HealthBench Evaluations))
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra | not stated | 64.7 | — | OpenAI | GPT-6 Astra System Card (Table 6. HealthBench Evaluations)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoreSource excerpt64.7Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row |
| 2 | GPT-5.6 Sol | not stated | 60.5 | — | OpenAI | GPT-6 Astra System Card (Table 6. HealthBench Evaluations)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoreSource excerpt60.5Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row |
| 3 | GPT-5.6 Terra | not stated | 57.7 | — | OpenAI | GPT-6 Astra System Card (Table 6. HealthBench Evaluations)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoreSource excerpt57.7Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row |
| 4 | GPT-5.6 Luna | not stated | 55.7 | — | OpenAI | GPT-6 Astra System Card (Table 6. HealthBench Evaluations)Retrieved 2026-11-12 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted scoreSource excerpt55.7Location Table 6 "HealthBench Evaluations - reported as length-adjusted score", HealthBench Professional row |
official results (Claude Fable 5.2 and Claude Mythos 5.2 System Card)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Max | 67.7 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt67.7%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 2 | Claude Sonnet 5.5 | Max | 63 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt63.0%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 3 | Claude Opus 5.5 | Max | 59.9 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback enabled. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt59.9%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
| 4 | Claude Haiku 5.5 | Max | 58.4 | — | Artificial Analysis | Claude Fable 5.2 and Claude Mythos 5.2 System CardRetrieved 2026-11-14 · Vendor-reported value. Independently run by Artificial Analysis at Anthropic’s request. max, length-adjusted score, GPT-5.4 low grader. fallback status is not stated for this model. Not comparable with internal runs in earlier Claude system cards (Opus 4.8 grader). Cost not stated.Source excerpt58.4%Location §8.16.2 “HealthBench Professional (Artificial Analysis)”, body text, the model’s length-adjusted score in parentheses |
official results (Addendum to GPT-6 Astra System Card: GPT-6.1 Terra)
| Rank | Model | effort | Score | Source cost (metric) | measured/reported by | source |
|---|---|---|---|---|---|---|
| 1 | GPT-6.1 Sol | not stated | 64.2 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt64.2 (67.2, 3038)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-1-sol column |
| 2 | GPT-6.1 Terra | not stated | 62.7 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt62.7 (64.2, 2531)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-1-terra column |
| 3 | GPT-6 Luna | not stated | 60.8 | — | OpenAI | Addendum to GPT-6 Astra System Card: GPT-6.1 TerraRetrieved 2026-11-14 · Vendor-reported value. Self-reported by OpenAI. effort not stated. length-adjusted score. Parentheses contain the unadjusted score and mean response length in characters; only the leading value is used as score. Cost not stated.Source excerpt60.8 (61.2, 2119)Location §5.1 Table 5, HealthBench Professional length-adjusted row, gpt-6-luna column |