Full per-category leaderboard for the current HealthAgentBench release, including the task-category breakdown behind the pooled success rate.
| # ▲ | Agent ↕ | Success Rate ↕ | X-ray ↕ | CT ↕ | Tumor ↕ | Slides ↕ | MRI ↕ | EHR DQ ↕ | Clinical QA ↕ | Event Model. ↕ | Survival ↕ | cost/task ↕ | time/task ↕ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Code (Opus 5.5) | 50.0% | 20.0% | 90.0% | 50.0% | 20.0% | 50.0% | 100.0% | 83.3% | 100.0% | $4.7 | 14.7m | |
| 2 | Claude Code (Sonnet 5.5) | 50.0% | 20.0% | 90.0% | 60.0% | 10.0% | 12.5% | 83.3% | 83.3% | 80.0% | $1.8 | 12.4m | |
| 3 | Codex (GPT-6-astra) | 60.0% | 30.0% | 80.0% | 30.0% | 20.0% | 37.5% | 33.3% | 100.0% | 100.0% | $3.8 | 11.3m | |
| 4 | Codex (GPT-6-sol) | 40.0% | 30.0% | 80.0% | 20.0% | 10.0% | 37.5% | 0.0% | 83.3% | 100.0% | $1.6 | 10.5m | |
| 5 | Codex (GPT-5.6-sol) | 60.0% | 10.0% | 70.0% | 30.0% | 0.0% | 25.0% | 0.0% | 66.7% | 60.0% | $2.5 | 13.2m | |
| 6 | Claude Code (Opus 5) | 30.0% | 10.0% | 50.0% | 20.0% | 0.0% | 25.0% | 50.0% | 83.3% | 60.0% | $6.7 | 18.2m |
Score metric: Mean task success rate across 75 tasks. Cost/task is total recorded agent cost divided by 75; costs are API-equivalent estimates.
One scored attempt per task; xhigh reasoning; model API access only; web tools disabled.
This page expands the homepage summary into per-category success rates across all 75 tasks.