Compare current model-and-harness results for SpatialBench-Long. Long-horizon agentic analysis of spatial biology datasets requiring multi-step biological reasoning. Current V1 covers 21 evaluations and 44 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Claude Code, LSVP, 40.5%); GPT-6.1 Terra (Pi, 40.5%); GPT-6.1 Sol (Codex, 39.9%). The highest-ranked published run without an LSVP exemption is GPT-6.1 Terra (Pi, 40.5%). 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-Long study reports 24 evaluations. This leaderboard reports the current V1 evaluation set of 21 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.
| Rank | Model | Harness | Provider | Score | Variant |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5* | Claude Code | Anthropic | 40.5% | LSVP |
| 2 | GPT-6.1 Terra | Pi | OpenAI | 40.5% | |
| 3 | GPT-6.1 Sol | Codex | OpenAI | 39.9% | |
| 4 | Claude Opus 5.5* | Pi | Anthropic | 39.3% | LSVP |
| 5 | GPT-6 Astra | Codex | OpenAI | 39.3% | |
| 6 | GPT-6.1 Terra | Codex | OpenAI | 39.3% | |
| 7 | GPT-6.1 Sol | Pi | OpenAI | 38.7% | |
| 8 | GPT-5.6 Sol | Pi | OpenAI | 37.5% | |
| 9 | GPT-6 Astra | Pi | OpenAI | 37.5% | |
| 10 | Claude Sonnet 5.5 | Claude Code | Anthropic | 36.5% | |
| 11 | GPT-5.6 Sol | Codex | OpenAI | 36.3% | |
| 12 | Claude Sonnet 5.5 | Pi | Anthropic | 34.9% | |
| 13 | Claude Opus 4.8 | Pi | Anthropic | 33.3% | |
| 14 | GPT-5.5 | Pi | OpenAI | 32.1% | |
| 15 | GPT-6 Sol | Pi | OpenAI | 31.6% | |
| 16 | Claude Opus 5 | Claude Code | Anthropic | 30.9% | |
| 17 | GPT-6 Sol | Codex | OpenAI | 30.9% | |
| 18 | Claude Fable 5.1* | Pi | Anthropic | 29.8% | LSVP |
| 19 | Claude Opus 4.7 | Claude Code | Anthropic | 29.8% | |
| 20 | Gemini 3.7 Flash | Pi | 29.8% | ||
| 21 | Gemini 3.8 Flash | Pi | 28.6% | ||
| 22 | Gemini 4 Argon | Pi | 28.6% | ||
| 23 | Claude Opus 5 | Pi | Anthropic | 28.0% | |
| 24 | Claude Opus 4.8 | Claude Code | Anthropic | 27.4% | |
| 25 | Grok 4.6 | Grok Build | SpaceXAI | 27.4% | |
| 26 | Claude Opus 4.7 | Pi | Anthropic | 26.2% | |
| 27 | DeepSeek V4.1 Flash | Pi | DeepSeek | 26.2% | |
| 28 | Grok 4.6 | Pi | SpaceXAI | 26.2% | |
| 29 | GPT-5.6 Luna | Codex | OpenAI | 25.0% | |
| 30 | Claude Sonnet 5 | Pi | Anthropic | 24.4% | |
| 31 | Grok 4.7 | Pi | SpaceXAI | 24.4% | |
| 32 | GPT-6 Luna | Pi | OpenAI | 23.2% | |
| 33 | Claude Sonnet 4.6 | Pi | Anthropic | 22.6% | |
| 34 | GPT-5.6 Luna | Pi | OpenAI | 22.6% | |
| 35 | Kimi K3 | Pi | Moonshot | 22.6% | |
| 36 | Mistral Large 4.0 | Pi | Mistral | 20.8% | |
| 37 | Claude Opus 5.5 | Pi | Anthropic | 20.2% | |
| 38 | Claude Sonnet 5 | Claude Code | Anthropic | 20.2% | |
| 39 | Claude Opus 5.5 | Claude Code | Anthropic | 19.6% | |
| 40 | GPT-6 Luna | Codex | OpenAI | 19.6% | |
| 41 | GLM 5.3 | Pi | Z.ai | 18.4% | |
| 42 | Claude Sonnet 4.6 | Claude Code | Anthropic | 17.3% | |
| 43 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 15.9% | |
| 44 | Nemotron 3.5 Lightning | Pi | NVIDIA | 4.8% |
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.