benchmarks.bio

Omics · Capabilities

EpiBench results

Agentic benchmark tasks drawn from real epigenomics analyses.

Published score: Current V1 evaluation set (106 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for EpiBench. Agentic benchmark tasks drawn from real epigenomics analyses. Current V1 covers 106 evaluations and 42 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Claude Code, LSVP, 39.6%); Claude Opus 5.5 (Pi, 39.0%); Claude Opus 5.5 (Claude Code, 38.7%). The highest-ranked published run without an LSVP exemption is Claude Opus 5.5 (Pi, 39.0%). Scores report benchmark-specific biological task performance on a 0–100 percentage scale. See the linked paper and repository for task construction, grading and uncertainty methodology. Results describe the complete model and agent harness used for each run, and are not comparable across different benchmarks or evaluation versions.

Published EpiBench scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5*Claude CodeAnthropic39.6%LSVP
2Claude Opus 5.5PiAnthropic39.0%
3Claude Opus 5.5Claude CodeAnthropic38.7%
4Claude Opus 5.5*PiAnthropic37.4%LSVP
5Claude Fable 5.1*PiAnthropic31.8%LSVP
6Claude Sonnet 5.5PiAnthropic30.2%
7Claude Sonnet 5.5Claude CodeAnthropic28.9%
8Gemini 4 ArgonPiGoogle28.9%
9GPT-5.5PiOpenAI28.3%
10GPT-6 AstraPiOpenAI28.3%
11GPT-6.1 SolPiOpenAI28.3%
12Claude Opus 4.8PiAnthropic25.8%
13GPT-5.6 SolCodexOpenAI25.8%
14DeepSeek V4.1 FlashPiDeepSeek25.5%
15Claude Opus 5PiAnthropic25.2%
16GPT-6.1 SolCodexOpenAI24.8%
17GPT-6 AstraCodexOpenAI24.2%
18Grok 4.6PiSpaceXAI23.6%
19GPT-6.1 TerraPiOpenAI23.3%
20Claude Opus 4.7PiAnthropic22.6%
21GPT-6.1 TerraCodexOpenAI22.6%
22Gemini 3.7 FlashPiGoogle22.0%
23GPT-6 SolPiOpenAI22.0%
24Claude Sonnet 5Claude CodeAnthropic21.7%
25Gemini 3.8 FlashPiGoogle21.4%
26GPT-5.6 SolPiOpenAI21.4%
27Grok 4.6Grok BuildSpaceXAI21.1%
28Claude Opus 5Claude CodeAnthropic20.8%
29Grok 4.7PiSpaceXAI20.8%
30Mistral Large 4.0PiMistral20.8%
31Claude Opus 4.8Claude CodeAnthropic20.4%
32GPT-5.6 LunaCodexOpenAI19.8%
33Claude Sonnet 5PiAnthropic19.5%
34GPT-6 LunaPiOpenAI19.2%
35Claude Opus 4.7Claude CodeAnthropic18.9%
36Claude Sonnet 4.6PiAnthropic18.9%
37GPT-5.6 LunaPiOpenAI18.9%
38Claude Sonnet 4.6Claude CodeAnthropic17.3%
39Kimi K3PiMoonshot15.7%
40GLM 5.3PiZ.ai12.9%
41Nemotron 3 Ultra 550B A55BPiNVIDIA11.3%
42Nemotron 3.5 LightningPiNVIDIA6.9%

Methodology and provenance

Scores use a 0–100 percentage scale. See Paper PDF, Original publication, 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.