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.
| Rank | Model | Harness | Provider | Score | Variant |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5* | Pi | Anthropic | 36.9% | LSVP |
| 2 | Claude Opus 5.5* | Claude Code | Anthropic | 32.4% | LSVP |
| 3 | Claude Sonnet 5.5 | Claude Code | Anthropic | 31.8% | |
| 4 | Claude Sonnet 5.5 | Pi | Anthropic | 30.3% | |
| 5 | GPT-5.6 Sol | Codex | OpenAI | 29.6% | |
| 6 | Gemini 4 Argon | Pi | 28.4% | ||
| 7 | GPT-6 Sol | Pi | OpenAI | 28.4% | |
| 8 | GPT-5.6 Sol | Pi | OpenAI | 27.8% | |
| 9 | Claude Opus 5 | Pi | Anthropic | 27.3% | |
| 10 | GPT-6 Astra | Pi | OpenAI | 27.3% | |
| 11 | GPT-6 Sol | Codex | OpenAI | 26.7% | |
| 12 | GPT-6.1 Sol | Pi | OpenAI | 26.7% | |
| 13 | Claude Fable 5.1* | Pi | Anthropic | 25.6% | LSVP |
| 14 | GPT-6 Astra | Codex | OpenAI | 25.0% | |
| 15 | GPT-6.1 Sol | Codex | OpenAI | 25.0% | |
| 16 | Claude Opus 5 | Claude Code | Anthropic | 24.4% | |
| 17 | GPT-5.5 | Pi | OpenAI | 21.6% | |
| 18 | Claude Opus 4.8 | Claude Code | Anthropic | 21.0% | |
| 19 | Gemini 3.8 Flash | Pi | 19.9% | ||
| 20 | Gemini 3.7 Flash | Pi | 18.8% | ||
| 21 | GPT-6.1 Terra | Codex | OpenAI | 18.8% | |
| 22 | GPT-6.1 Terra | Pi | OpenAI | 18.2% | |
| 23 | Claude Opus 4.8 | Pi | Anthropic | 16.5% | |
| 24 | Grok 4.6 | Pi | SpaceXAI | 16.5% | |
| 25 | Claude Sonnet 5 | Pi | Anthropic | 15.9% | |
| 26 | Claude Sonnet 5 | Claude Code | Anthropic | 15.9% | |
| 27 | Grok 4.6 | Grok Build | SpaceXAI | 15.9% | |
| 28 | DeepSeek V4.1 Flash | Pi | DeepSeek | 14.8% | |
| 29 | Kimi K3 | Pi | Moonshot | 14.2% | |
| 30 | Claude Opus 5.5 | Claude Code | Anthropic | 13.6% | |
| 31 | GPT-5.6 Luna | Pi | OpenAI | 13.1% | |
| 32 | GPT-6 Luna | Codex | OpenAI | 11.9% | |
| 33 | Claude Opus 4.7 | Claude Code | Anthropic | 11.4% | |
| 34 | Claude Opus 4.7 | Pi | Anthropic | 11.4% | |
| 35 | Grok 4.7 | Pi | SpaceXAI | 11.4% | |
| 36 | GPT-6 Luna | Pi | OpenAI | 10.8% | |
| 37 | Claude Sonnet 4.6 | Claude Code | Anthropic | 10.2% | |
| 38 | GLM 5.3 | Pi | Z.ai | 10.2% | |
| 39 | Mistral Large 4.0 | Pi | Mistral | 10.2% | |
| 40 | GPT-5.6 Luna | Codex | OpenAI | 9.7% | |
| 41 | Claude Sonnet 4.6 | Pi | Anthropic | 8.5% | |
| 42 | Claude Opus 5.5 | Pi | Anthropic | 6.8% | |
| 43 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 3.0% | |
| 44 | Nemotron 3.5 Lightning | Pi | NVIDIA | 0.0% |
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.