Benchmark Radar™
Star —

LatchBio SpatialBench

Explore score history ↗

159 spatial transcriptomics analysis evaluations, each run 3 times and averaged; the published 95% intervals span about ±6 to 7 points, so neighbouring models usually overlap. The score depends on the agent harness as well as the model (Claude Opus 4.8 scores 55.35 in Claude Code and 52.62 in mini-swe-agent), and the evaluation container has network access. The arXiv paper (2512.21907) reports an earlier task set, so its numbers are not comparable with these.

Evaluates
biology
Openness
unknown
Importer
Model reports

LatchBio SpatialBench code

LatchBio SpatialBench results and reported scores

Reported benchmark scores · Model reports

Reported scores: 20

ModelOrganizationReported valueReportedEvidence
Claude Opus 4.5Anthropic42.772026-11-01evidence
Claude Opus 4.6Anthropic52.832026-11-01evidence
Claude Opus 4.7Anthropic51.362026-11-01evidence
Claude Opus 4.7Anthropic52.412026-11-01evidence
Claude Opus 4.8Anthropic55.352026-11-01evidence
Claude Opus 4.8Anthropic52.622026-11-01evidence
Claude Sonnet 4.5Anthropic41.512026-11-01evidence
Claude Sonnet 4.6Anthropic44.232026-11-01evidence
Gemini 2.5 ProGoogle28.932026-11-01evidence
Gemini 3.1 Pro PreviewGoogle51.572026-11-01evidence
Gemini 3.5 FlashGoogle48.852026-11-01evidence
Gemini 3.5 FlashGoogle55.562026-11-01evidence
GPT-5.1OpenAI39.832026-11-01evidence
GPT-5.2OpenAI50.12026-11-01evidence
GPT-5.4OpenAI57.442026-11-01evidence
GPT-5.5OpenAI53.672026-11-01evidence
GPT-5.5OpenAI57.652026-11-01evidence
Grok 4xAI31.872026-11-01evidence
Grok 4.1 Fast (reasoning)xAI33.962026-11-01evidence
Grok 4.20 Beta 0309 (reasoning)xAI45.912026-11-01evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.

Which sources cite LatchBio SpatialBench?

Related biology benchmarks