Results

Compare model results reported by benchmark papers and official leaderboards. BioInspect reruns are attached to their individual evaluation records.

Source-reported scores are shown as published and are not combined into a single index because benchmarks may use different metrics, model versions, prompts, and harnesses.

Results within one benchmark

GPT-5.6 Sol
62.1%
Opus 5
60.1%
Claude Opus 4.6
52.8%
Claude Opus 4.5
49.9%
GPT-5.2
45.2%
Claude Sonnet 4.5
44.2%
GPT-5.1
37.9%
Grok-4.1
35.6%
Grok-4
33.9%
Gemini 2.5 Pro
29.2%

Cross-benchmark results matrix

Source-reported results are shown without aggregation. Blank cells indicate that the source did not report that model on the benchmark. Scores may use different metrics, variants, and evaluation harnesses and should not be treated as directly comparable across columns.

ModelscBench: A Benchmark for Single-Cell RNA-seq AnalysisSpatialBench: Can Agents Analyze Real-World Spatial Biology Data?BiomniBench-DataAnalysis (Biomni-DA-v0)BixBenchLifeSciBench: Realistic, Expert-Level Life Science TasksSpatialBench-Long: Verifiable Long-Horizon Spatial BiologyBixBench-Verified-50GeneBench-Pro: Multistage Statistical Reasoning in Genomics & BiologyscBench-Long: Verifiable Long-Horizon Single-Cell BiologyPubMedQA: A Dataset for Biomedical Research Question Answering
Biomni Lab (2026-02-03)65.0%52.2%88.7%
Claude Code (Opus 4.6)47.8%39.5%65.3%
Claude Opus 4.549.9%38.4%77.5%
Claude Opus 4.652.8%
Claude Opus 4.816.0%25.4%
Claude Sonnet 4.544.2%28.3%0.0%
Edison Analysis52.7%42.4%78.0%
GPT-4o / Claude 3.5 Sonnet17.0%
GPT-5.137.9%27.4%
GPT-5.245.2%34.0%
GPT-5.447.9%
GPT-5.551.9%11.1%
GPT-5.6 Sol62.1%74.5%38.9%28.7%38.1%
GPT-5.6 Sol Pro31.5%
GPT-Rosalind57.6%
Gemini 2.5 Pro29.2%20.1%
Gemini 3.1 Pro51.5%
Gemini 3.5 Flash11.1%8.1%22.2%
Grok 4.339.9%
Grok-433.9%22.8%
Grok-4.135.6%24.7%
Junior scientists48.5%
OpenAI Agents SDK (GPT-5.2)51.4%38.5%61.3%
Opus 560.1%71.0%25.0%41.3%
Senior pharmaceutical scientists68.5%