PhageBench - Contamination Detection: leaderboard
Deciding whether a phage contig carries host-derived sequence from raw bacteriophage genome sequences on PhageBench (2026), multiple-choice questions built from curated phage databases and answered zero-shot with chain-of-thought prompting. Accuracy in percent; higher is better.
Metric: Accuracy (%) on 1,200 two-option items deciding whether a phage contig carries host-derived sequence, each a raw-nucleotide multiple-choice question with shuffled options, zero-shot chain-of-thought prompting (reasoning models at reasoning effort medium with at most 2,048 reasoning tokens); higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 8 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3 Flash | Medium | 58.83 | #175 |
| 2 | GPT-5.2 | Medium | 54.17 | #185 |
| 3 | Qwen 3 Max | Thinking | 52.58 | #415 |
| 4 | Claude Sonnet 4.5 | Thinking | 51.83 | #261 |
| 5 | GPT-4o Mini | 50.08 | #1667 | |
| 6 | LLaMA 4 (PhageBench checkpoint unspecified) | 49.58 | ||
| 7 | Qwen3-235B (PhageBench checkpoint unspecified) | 49.33 | ||
| 8 | GPT-OSS-120B | 46.58 | #704 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=phagebench-contamination-detection · How It Works · Data refreshed daily, snapshot 2026-11-21.