Compare current model-and-harness results for VariantBench. Agentic genetic variant discovery and interpretation tasks grounded in real biological analyses. Current V1 covers 118 evaluations and 42 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5 (Claude Code, 52.8%); Claude Sonnet 5.5 (Pi, 51.7%); Claude Sonnet 5.5 (Claude Code, 51.1%). 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.
| Rank | Model | Harness | Provider | Score | Variant |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | Claude Code | Anthropic | 52.8% | |
| 2 | Claude Sonnet 5.5 | Pi | Anthropic | 51.7% | |
| 3 | Claude Sonnet 5.5 | Claude Code | Anthropic | 51.1% | |
| 4 | Claude Opus 5 | Pi | Anthropic | 50.6% | |
| 5 | Claude Opus 5.5 | Pi | Anthropic | 50.6% | |
| 6 | Gemini 4 Argon | Pi | 50.3% | ||
| 7 | Claude Opus 5.5 | Claude Code | Anthropic | 49.7% | |
| 8 | Claude Opus 5.5* | Claude Code | Anthropic | 48.9% | LSVP |
| 9 | Claude Opus 5.5* | Pi | Anthropic | 48.6% | LSVP |
| 10 | GPT-6 Astra | Pi | OpenAI | 46.6% | |
| 11 | GPT-6.1 Sol | Pi | OpenAI | 46.0% | |
| 12 | Grok 4.7 | Pi | SpaceXAI | 45.5% | |
| 13 | GPT-6 Astra | Codex | OpenAI | 45.2% | |
| 14 | GPT-6.1 Sol | Codex | OpenAI | 43.2% | |
| 15 | Claude Opus 4.8 | Claude Code | Anthropic | 42.4% | |
| 16 | Claude Opus 4.8 | Pi | Anthropic | 42.4% | |
| 17 | Claude Fable 5.1* | Pi | Anthropic | 40.7% | LSVP |
| 18 | Grok 4.6 | Pi | SpaceXAI | 40.4% | |
| 19 | Claude Sonnet 5 | Pi | Anthropic | 39.8% | |
| 20 | Claude Sonnet 5 | Claude Code | Anthropic | 39.0% | |
| 21 | Grok 4.6 | Grok Build | SpaceXAI | 39.0% | |
| 22 | GPT-6 Sol | Pi | OpenAI | 38.7% | |
| 23 | GPT-5.6 Sol | Pi | OpenAI | 35.3% | |
| 24 | Claude Opus 4.7 | Pi | Anthropic | 35.0% | |
| 25 | GPT-6.1 Terra | Codex | OpenAI | 34.8% | |
| 26 | GPT-5.6 Sol | Codex | OpenAI | 34.2% | |
| 27 | GPT-6.1 Terra | Pi | OpenAI | 33.6% | |
| 28 | Claude Opus 4.7 | Claude Code | Anthropic | 33.0% | |
| 29 | Gemini 3.7 Flash | Pi | 31.4% | ||
| 30 | Gemini 3.8 Flash | Pi | 31.4% | ||
| 31 | Kimi K3 | Pi | Moonshot | 31.4% | |
| 32 | Claude Sonnet 4.6 | Pi | Anthropic | 31.1% | |
| 33 | DeepSeek V4.1 Flash | Pi | DeepSeek | 31.1% | |
| 34 | GPT-5.5 | Pi | OpenAI | 30.5% | |
| 35 | Mistral Large 4.0 | Pi | Mistral | 28.0% | |
| 36 | GPT-6 Luna | Pi | OpenAI | 27.7% | |
| 37 | GPT-5.6 Luna | Codex | OpenAI | 27.1% | |
| 38 | Claude Sonnet 4.6 | Claude Code | Anthropic | 26.3% | |
| 39 | GLM 5.3 | Pi | Z.ai | 26.3% | |
| 40 | GPT-5.6 Luna | Pi | OpenAI | 24.0% | |
| 41 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 17.8% | |
| 42 | Nemotron 3.5 Lightning | Pi | NVIDIA | 14.4% |
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.