<!-- Generated by scripts/build-backrooms-results-table.mjs. Do not edit directly; see docs/updating-results.md. -->

# OncologyBench results

> OncologyBench tests whether AI agents can translate cancer research into real-world decisions about detection, diagnosis, treatment, and research priorities. Separate agents complete analytical tasks, and a final agent synthesizes their findings into a decision. Strict scores require full marks on both tasks and decisions across the selected 25 evaluations; refusals and timed-out slots remain in the denominator.

Data updated: 2026-10-09

Artifact generated: 2026-10-13T15:43:09.595Z

- Domain: Applications
- Horizon: long
- Assessment: Capabilities
- Published score scope: Selected release (strict task and decision pass) (25 evaluations)

Scores use a 0–100 percentage scale. Consult the repository and paper for benchmark-specific task construction, grading, aggregation, and uncertainty methodology. Results compare model-and-harness combinations.

| Rank | Model | Harness | Provider | Score |
| ---: | --- | --- | --- | ---: |
| 1 | Claude Opus 5 | Pi | Anthropic | 40.0% |
| 2 | Claude Opus 5.5 | Claude Code | Anthropic | 37.3% |
| 3 | Claude Opus 5 | Claude Code | Anthropic | 36.0% |
| 4 | Claude Sonnet 5.5 | Claude Code | Anthropic | 32.0% |
| 5 | Claude Sonnet 5.5 | Pi | Anthropic | 29.3% |
| 6 | Claude Opus 5.5 | Pi | Anthropic | 29.3% |
| 7 | Kimi K3 | Pi | Moonshot | 24.0% |
| 8 | GPT-6 Astra | Pi | OpenAI | 21.3% |
| 9 | GPT-6 Sol | Codex | OpenAI | 20.0% |
| 10 | GPT-5.6 Sol | Pi | OpenAI | 18.7% |
| 11 | GPT-6 Astra | Codex | OpenAI | 18.7% |
| 12 | Grok 4.6 | Pi | SpaceXAI | 18.7% |
| 13 | GPT-6 Sol | Pi | OpenAI | 18.7% |
| 14 | Grok 4.7 | Grok Build | SpaceXAI | 18.7% |
| 15 | Grok 4.7 | Pi | SpaceXAI | 18.7% |
| 16 | Grok 4.6 | Grok Build | SpaceXAI | 17.3% |
| 17 | Claude Opus 4.8 | Claude Code | Anthropic | 16.0% |
| 18 | Claude Opus 4.8 | Pi | Anthropic | 16.0% |
| 19 | GPT-6 Luna | Codex | OpenAI | 16.0% |
| 20 | Claude Sonnet 5 | Pi | Anthropic | 14.7% |
| 21 | DeepSeek V4.1 Flash | Pi | DeepSeek | 14.7% |
| 22 | GPT-5.5 | Pi | OpenAI | 14.7% |
| 23 | GPT-5.6 Sol | Codex | OpenAI | 14.7% |
| 24 | Gemini 3.7 Flash | Pi | Google | 14.7% |
| 25 | Gemini 3.8 Flash | Pi | Google | 13.3% |
| 26 | Claude Sonnet 5 | Claude Code | Anthropic | 12.0% |
| 27 | GPT-6 Luna | Pi | OpenAI | 12.0% |
| 28 | GPT-5.6 Luna | Codex | OpenAI | 10.7% |
| 29 | GPT-5.6 Luna | Pi | OpenAI | 10.7% |
| 30 | Claude Sonnet 4.6 | Claude Code | Anthropic | 5.3% |
| 31 | Claude Sonnet 4.6 | Pi | Anthropic | 2.7% |

## Sources and downloads

Publication links are not yet available.
- [Interactive benchmark](https://benchmarks.bio/oncology)
- [JSON](https://benchmarks.bio/benchmarks/oncologybench/latest.json)
- [CSV](https://benchmarks.bio/benchmarks/oncologybench/latest.csv)
- [Legacy V0 cross-benchmark leaderboard](https://benchmarks.bio/results/v0/)