<!-- Generated by scripts/build-backrooms-results-table.mjs. Do not edit directly; see docs/updating-results.md. -->

# MetagenomicsBench results

> Agentic microbial community analysis tasks spanning shotgun, amplicon, long-read and spatially integrated metagenomics.

Data updated: 2026-10-09

Artifact generated: 2026-10-13T15:43:09.595Z

- Domain: Omics
- Horizon: short
- Assessment: Capabilities
- Published score scope: Full benchmark (100 evaluations)

Scores use a 0–100 percentage scale. Consult the repository and paper for benchmark-specific task construction, grading, aggregation, and uncertainty methodology. Results compare model-and-harness combinations.

| Rank | Model | Harness | Provider | Score |
| ---: | --- | --- | --- | ---: |
| 1 | Claude Sonnet 5.5 | Claude Code | Anthropic | 60.0% |
| 2 | Claude Sonnet 5.5 | Pi | Anthropic | 60.0% |
| 3 | Claude Opus 5 | Pi | Anthropic | 54.0% |
| 4 | Claude Opus 5 | Claude Code | Anthropic | 53.3% |
| 5 | Claude Opus 4.8 | Claude Code | Anthropic | 50.0% |
| 6 | GPT-6 Astra | Codex | OpenAI | 49.7% |
| 7 | Grok 4.7 | Grok Build | SpaceXAI | 48.7% |
| 8 | Claude Opus 4.8 | Pi | Anthropic | 48.3% |
| 9 | Grok 4.7 | Pi | SpaceXAI | 48.3% |
| 10 | GPT-6 Astra | Pi | OpenAI | 47.0% |
| 11 | Claude Opus 5.5 | Claude Code | Anthropic | 43.7% |
| 12 | Claude Sonnet 5 | Pi | Anthropic | 42.3% |
| 13 | Grok 4.6 | Grok Build | SpaceXAI | 42.0% |
| 14 | GPT-6 Sol | Pi | OpenAI | 42.0% |
| 15 | Claude Sonnet 5 | Claude Code | Anthropic | 41.0% |
| 16 | GPT-6 Sol | Codex | OpenAI | 40.7% |
| 17 | Claude Opus 5.5 | Pi | Anthropic | 39.3% |
| 18 | Grok 4.6 | Pi | SpaceXAI | 38.7% |
| 19 | DeepSeek V4.1 Flash | Pi | DeepSeek | 38.0% |
| 20 | Claude Opus 4.7 | Pi | Anthropic | 35.3% |
| 21 | Gemini 3.7 Flash | Pi | Google | 31.7% |
| 22 | GPT-5.6 Sol | Pi | OpenAI | 31.3% |
| 23 | GPT-6 Luna | Codex | OpenAI | 31.0% |
| 24 | GPT-6 Luna | Pi | OpenAI | 30.7% |
| 25 | GPT-5.6 Sol | Codex | OpenAI | 28.7% |
| 26 | Claude Sonnet 4.6 | Claude Code | Anthropic | 28.3% |
| 27 | Claude Sonnet 4.6 | Pi | Anthropic | 27.7% |
| 28 | Kimi K3 | Pi | Moonshot | 27.3% |
| 29 | Gemini 3.8 Flash | Pi | Google | 26.7% |
| 30 | GPT-5.6 Luna | Codex | OpenAI | 24.3% |
| 31 | GPT-5.6 Luna | Pi | OpenAI | 22.7% |

## Sources and downloads

- [Paper PDF](/papers/metagenomics.pdf)
- [Original publication](https://latch.bio/metagenomicsbench)
- [Repository](https://github.com/latchbio/metagenomicsbench)
- [Interactive benchmark](https://benchmarks.bio/metagenomics)
- [JSON](https://benchmarks.bio/benchmarks/metagenomics/latest.json)
- [CSV](https://benchmarks.bio/benchmarks/metagenomics/latest.csv)
- [Legacy V0 cross-benchmark leaderboard](https://benchmarks.bio/results/v0/)