This page reports published NeuroscienceBench scores from benchmarks.bio. NeuroscienceBench contains 91 evaluations built from real experimental data, each one graded deterministically against the biological result the original analysis reached, so a run counts as a pass only when the agent recovers that result rather than when it produces plausible-looking output. Results are reported per model and per harness rather than per model alone, because the harness an agent runs under changes its score substantially. Scores use a 0-100 percentage scale and are not comparable across benchmarks; see the linked repository and paper for task construction, grading, and confidence-interval methodology.
| Rank | Model | Harness | Provider | Score | Variant |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5* | Pi | Anthropic | 32.2% | LSVP |
| 2 | Claude Opus 5* | Pi | Anthropic | 30.8% | LSVP |
| 3 | Claude Opus 5.5* | Claude Code | Anthropic | 29.7% | LSVP |
| 4 | Claude Sonnet 5.5* | Claude Code | Anthropic | 29.7% | LSVP |
| 5 | Claude Fable 5.1* | Pi | Anthropic | 28.4% | LSVP |
| 6 | Grok 4.7 | Grok Build | SpaceXAI | 27.5% | |
| 7 | Claude Sonnet 5.5* | Pi | Anthropic | 27.1% | LSVP |
| 8 | Claude Opus 5* | Claude Code | Anthropic | 25.6% | LSVP |
| 9 | Claude Fable 5.1* | Claude Code | Anthropic | 23.8% | LSVP |
| 10 | Grok 4.7 | Pi | SpaceXAI | 23.4% | |
| 11 | GPT-6 Astra | Pi | OpenAI | 22.7% | |
| 12 | GPT-6.1 Terra | Pi | OpenAI | 22.3% | |
| 13 | GPT-6.1 Sol | Pi | OpenAI | 22.0% | |
| 14 | DeepSeek V4.1 Flash | Pi | DeepSeek | 22.0% | |
| 15 | GPT-6 Luna | Codex | OpenAI | 21.5% | |
| 16 | Grok 4.6 | Grok Build | SpaceXAI | 21.2% | |
| 17 | GPT-6 Astra | Codex | OpenAI | 20.5% | |
| 18 | Grok 4.6 | Pi | SpaceXAI | 20.5% | |
| 19 | GPT-6 Luna | Pi | OpenAI | 20.4% | |
| 20 | GPT-6.1 Sol | Codex | OpenAI | 20.1% | |
| 21 | GPT-6.1 Terra | Codex | OpenAI | 20.1% | |
| 22 | Claude Sonnet 5* | Pi | Anthropic | 19.4% | LSVP |
| 23 | Mistral Large 4.0 | Pi | Mistral | 19.4% | |
| 24 | Gemini 4 Argon | Pi | 19.1% | ||
| 25 | Claude Sonnet 5* | Claude Code | Anthropic | 18.7% | LSVP |
| 26 | GLM 5.3 | Pi | Z.ai | 16.5% | |
| 27 | Kimi K3 | Pi | Moonshot | 16.5% | |
| 28 | Gemini 3.8 Flash | Pi | 15.8% | ||
| 29 | GPT-5.6 Sol | Codex | OpenAI | 15.8% | |
| 30 | GPT-5.6 Luna | Pi | OpenAI | 15.0% | |
| 31 | GPT-5.6 Sol | Pi | OpenAI | 13.9% | |
| 32 | GPT-5.6 Luna | Codex | OpenAI | 12.5% | |
| 33 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 12.1% | |
| 34 | Nemotron 3.5 Lightning | Pi | NVIDIA | 7.7% |
Methodology and provenance
Scores use a 0–100 percentage scale. See Repository for task construction, grading, aggregation, and uncertainty details.
LSVP: Anthropic's safety-filter-exempt evaluation variant, which produces fewer refusals on biology prompts. LSVP results are reported alongside the model's standard run and are not a replacement for it.