benchmarks.bio

Omics · Capabilities

SpatialBench results

Agentic analysis of real spatial transcriptomics datasets across platforms and biological task types.

Published score: Current V1 evaluation set (115 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for SpatialBench. Agentic analysis of real spatial transcriptomics datasets across platforms and biological task types. Also known as SpatialBench Verified. Current V1 covers 115 evaluations and 43 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are GPT-6.1 Terra (Pi, 73.3%); Claude Opus 5.5 (Pi, 72.8%); Claude Sonnet 5.5 (Pi, 72.8%). 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 SpatialBench study reports 146 evaluations. This leaderboard reports the current V1 evaluation set of 115 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 SpatialBench scores
RankModelHarnessProviderScoreVariant
1GPT-6.1 TerraPiOpenAI73.3%
2Claude Opus 5.5PiAnthropic72.8%
3Claude Sonnet 5.5PiAnthropic72.8%
4Muse Wonder 1.0PiMeta71.9%
5GPT-6.1 TerraCodexOpenAI70.7%
6Claude Sonnet 5.5Claude CodeAnthropic70.1%
7Claude Opus 5.5Claude CodeAnthropic69.6%
8Grok 4.7PiSpaceXAI69.6%
9Claude Opus 5.5*PiAnthropic69.0%LSVP
10Gemini 4 ArgonPiGoogle69.0%
11Claude Opus 5.5*Claude CodeAnthropic68.7%LSVP
12Claude Opus 5PiAnthropic68.4%
13Claude Opus 5Claude CodeAnthropic68.4%
14GPT-5.6 SolCodexOpenAI68.1%
15Claude Fable 5.1*PiAnthropic67.5%LSVP
16GPT-5.6 SolPiOpenAI67.5%
17GPT-6 AstraCodexOpenAI67.5%
18GPT-6 SolPiOpenAI67.5%
19GPT-6.1 SolPiOpenAI67.3%
20Grok 4.6PiSpaceXAI67.3%
21Claude Opus 4.8Claude CodeAnthropic67.0%
22GPT-6.1 SolCodexOpenAI67.0%
23Claude Sonnet 5PiAnthropic66.7%
24GPT-6 AstraPiOpenAI66.7%
25Grok 4.6Grok BuildSpaceXAI66.1%
26Claude Opus 4.8PiAnthropic65.8%
27Claude Sonnet 5Claude CodeAnthropic65.8%
28GPT-5.5PiOpenAI64.1%
29Kimi K3PiMoonshot63.5%
30DeepSeek V4.1 FlashPiDeepSeek63.2%
31GPT-5.6 LunaPiOpenAI60.9%
32GPT-6 LunaPiOpenAI59.7%
33Gemini 3.8 FlashPiGoogle59.4%
34Gemini 3.7 FlashPiGoogle59.1%
35GPT-5.6 LunaCodexOpenAI58.5%
36GLM 5.3PiZ.ai57.4%
37Claude Opus 4.7Claude CodeAnthropic56.5%
38Mistral Large 4.0PiMistral55.9%
39Claude Opus 4.7PiAnthropic54.8%
40Claude Sonnet 4.6PiAnthropic53.9%
41Claude Sonnet 4.6Claude CodeAnthropic53.3%
42Nemotron 3 Ultra 550B A55BPiNVIDIA47.0%
43Nemotron 3.5 LightningPiNVIDIA29.0%

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.