Compare current model-and-harness results for scBench. Agentic analysis of real single-cell RNA-seq datasets across platforms and task categories. Also known as SingleCellBench. Current V1 covers 195 evaluations and 42 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are GPT-6 Astra (Pi, 63.9%); GPT-6.1 Sol (Codex, 63.6%); GPT-6 Astra (Codex, 63.4%). 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 study reports 394 evaluations. This leaderboard reports the current V1 evaluation set of 195 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 | GPT-6 Astra | Pi | OpenAI | 63.9% | |
| 2 | GPT-6.1 Sol | Codex | OpenAI | 63.6% | |
| 3 | GPT-6 Astra | Codex | OpenAI | 63.4% | |
| 4 | GPT-6.1 Sol | Pi | OpenAI | 62.9% | |
| 5 | Claude Opus 5.5* | Pi | Anthropic | 61.9% | LSVP |
| 6 | Claude Opus 5.5* | Claude Code | Anthropic | 61.5% | LSVP |
| 7 | GPT-6 Sol | Pi | OpenAI | 61.5% | |
| 8 | Claude Fable 5.1* | Pi | Anthropic | 60.5% | LSVP |
| 9 | Claude Sonnet 5.5 | Pi | Anthropic | 60.3% | |
| 10 | GPT-5.6 Sol | Codex | OpenAI | 59.8% | |
| 11 | Claude Sonnet 5.5 | Claude Code | Anthropic | 59.1% | |
| 12 | Gemini 4 Argon | Pi | 59.1% | ||
| 13 | Grok 4.6 | Pi | SpaceXAI | 59.1% | |
| 14 | GPT-5.6 Sol | Pi | OpenAI | 58.8% | |
| 15 | Grok 4.7 | Pi | SpaceXAI | 58.5% | |
| 16 | Claude Opus 5 | Pi | Anthropic | 58.3% | |
| 17 | GPT-6.1 Terra | Pi | OpenAI | 58.3% | |
| 18 | GPT-6.1 Terra | Codex | OpenAI | 58.3% | |
| 19 | Claude Opus 5 | Claude Code | Anthropic | 57.6% | |
| 20 | Grok 4.6 | Grok Build | SpaceXAI | 57.6% | |
| 21 | Claude Opus 4.8 | Pi | Anthropic | 57.1% | |
| 22 | GPT-5.5 | Pi | OpenAI | 57.1% | |
| 23 | Claude Opus 5.5 | Claude Code | Anthropic | 56.4% | |
| 24 | Claude Opus 4.8 | Claude Code | Anthropic | 56.2% | |
| 25 | Claude Sonnet 5 | Pi | Anthropic | 55.6% | |
| 26 | Claude Opus 5.5 | Pi | Anthropic | 55.4% | |
| 27 | DeepSeek V4.1 Flash | Pi | DeepSeek | 54.7% | |
| 28 | Claude Sonnet 5 | Claude Code | Anthropic | 53.9% | |
| 29 | GPT-6 Luna | Pi | OpenAI | 53.9% | |
| 30 | Mistral Large 4.0 | Pi | Mistral | 53.0% | |
| 31 | Claude Opus 4.7 | Claude Code | Anthropic | 52.5% | |
| 32 | Claude Opus 4.7 | Pi | Anthropic | 50.8% | |
| 33 | Gemini 3.7 Flash | Pi | 50.4% | ||
| 34 | GPT-5.6 Luna | Pi | OpenAI | 50.3% | |
| 35 | GPT-5.6 Luna | Codex | OpenAI | 48.9% | |
| 36 | GLM 5.3 | Pi | Z.ai | 48.7% | |
| 37 | Kimi K3 | Pi | Moonshot | 47.5% | |
| 38 | Claude Sonnet 4.6 | Claude Code | Anthropic | 41.7% | |
| 39 | Claude Sonnet 4.6 | Pi | Anthropic | 39.5% | |
| 40 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 36.6% | |
| 41 | Gemini 3.8 Flash | Pi | 34.0% | ||
| 42 | Nemotron 3.5 Lightning | Pi | NVIDIA | 22.9% |
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.