benchmarks.bio

Omics · Capabilities

scBench-Long results

Long-horizon single-cell analysis tasks requiring sustained empirical work and biological reasoning.

Published score: Current V1 evaluation set (22 evaluations) · long horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for scBench-Long. Long-horizon single-cell analysis tasks requiring sustained empirical work and biological reasoning. Current V1 covers 22 evaluations and 44 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Pi, LSVP, 36.9%); Claude Opus 5.5* (Claude Code, LSVP, 32.4%); Claude Sonnet 5.5 (Claude Code, 31.8%). The highest-ranked published run without an LSVP exemption is Claude Sonnet 5.5 (Claude Code, 31.8%). 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.

Original study and current evaluation set

The original scBench-Long study reports 21 evaluations. This leaderboard reports the current V1 evaluation set of 22 evaluations. These are separately scoped releases. Use the current V1 set when citing this table and the original study when citing its historical results. Score comparisons require the same evaluation version and scoring rules. Original study.

Published scBench-Long scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5*PiAnthropic36.9%LSVP
2Claude Opus 5.5*Claude CodeAnthropic32.4%LSVP
3Claude Sonnet 5.5Claude CodeAnthropic31.8%
4Claude Sonnet 5.5PiAnthropic30.3%
5GPT-5.6 SolCodexOpenAI29.6%
6Gemini 4 ArgonPiGoogle28.4%
7GPT-6 SolPiOpenAI28.4%
8GPT-5.6 SolPiOpenAI27.8%
9Claude Opus 5PiAnthropic27.3%
10GPT-6 AstraPiOpenAI27.3%
11GPT-6 SolCodexOpenAI26.7%
12GPT-6.1 SolPiOpenAI26.7%
13Claude Fable 5.1*PiAnthropic25.6%LSVP
14GPT-6 AstraCodexOpenAI25.0%
15GPT-6.1 SolCodexOpenAI25.0%
16Claude Opus 5Claude CodeAnthropic24.4%
17GPT-5.5PiOpenAI21.6%
18Claude Opus 4.8Claude CodeAnthropic21.0%
19Gemini 3.8 FlashPiGoogle19.9%
20Gemini 3.7 FlashPiGoogle18.8%
21GPT-6.1 TerraCodexOpenAI18.8%
22GPT-6.1 TerraPiOpenAI18.2%
23Claude Opus 4.8PiAnthropic16.5%
24Grok 4.6PiSpaceXAI16.5%
25Claude Sonnet 5PiAnthropic15.9%
26Claude Sonnet 5Claude CodeAnthropic15.9%
27Grok 4.6Grok BuildSpaceXAI15.9%
28DeepSeek V4.1 FlashPiDeepSeek14.8%
29Kimi K3PiMoonshot14.2%
30Claude Opus 5.5Claude CodeAnthropic13.6%
31GPT-5.6 LunaPiOpenAI13.1%
32GPT-6 LunaCodexOpenAI11.9%
33Claude Opus 4.7Claude CodeAnthropic11.4%
34Claude Opus 4.7PiAnthropic11.4%
35Grok 4.7PiSpaceXAI11.4%
36GPT-6 LunaPiOpenAI10.8%
37Claude Sonnet 4.6Claude CodeAnthropic10.2%
38GLM 5.3PiZ.ai10.2%
39Mistral Large 4.0PiMistral10.2%
40GPT-5.6 LunaCodexOpenAI9.7%
41Claude Sonnet 4.6PiAnthropic8.5%
42Claude Opus 5.5PiAnthropic6.8%
43Nemotron 3 Ultra 550B A55BPiNVIDIA3.0%
44Nemotron 3.5 LightningPiNVIDIA0.0%

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.