Atlas

Benchmarks

← All benchmarks

Biopharma Bench V0.1 (Pass@1)

Health & Medicine · 2026-09-22

The share of Biopharma Bench V0.1's 71 pharmaceutical regulatory and CMC tasks on which an agent's deliverable passes every one of the task's frozen expert criteria in its single scored trial, in percent. It is the strict counterpart of the mean score, which gives credit for each criterion passed; the same LLM rubric grader and deterministic checks decide every criterion, against the V0.1 rubrics.

Benchmark source: Raycaster

Top models (higher is better)

ModelScoreSource
Claude Fable 5.111.3Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
GPT-6 Astra11.3Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Claude Opus 5.55.6Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Claude Opus 52.8Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Grok 4.62.8Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Grok 4.72.8Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
GPT-5.6 Sol1.4Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Kimi K31.4Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
DeepSeek V4.1 Flash0.0Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Gemini 3.8 Flash0.0Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
GLM-5.3 Flash0.0Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
GPT-6 Sol0.0Biopharma Bench V0.1 results board (snapshot 2026-11-02) · Raycaster
Loading Atlas data…