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

# TxBench-Antibody-Discovery results

> Agentic benchmark tasks for antibody discovery and development decisions.

Data updated: 2026-09-20

Artifact generated: 2026-09-21T21:43:49.599Z

- Domain: Therapeutics
- 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 | GPT-6 Astra | Pi | OpenAI | 56.9% |
| 2 | GPT-6 Astra | Codex | OpenAI | 55.2% |
| 3 | Claude Opus 5 | Claude Code | Anthropic | 53.0% |
| 4 | Grok 4.6 | Pi | SpaceXAI | 51.9% |
| 5 | Claude Opus 5 | Pi | Anthropic | 51.4% |
| 6 | Grok 4.6 | Grok Build | SpaceXAI | 51.0% |
| 7 | Claude Opus 4.8 | Pi | Anthropic | 45.6% |
| 8 | Gemini 3.5 Flash | Pi | Google | 42.6% |
| 9 | Claude Opus 4.8 | Claude Code | Anthropic | 42.0% |
| 10 | Claude Opus 4.7 | Claude Code | Anthropic | 39.0% |
| 11 | DeepSeek V4.1 Flash | Pi | DeepSeek | 39.0% |
| 12 | Grok 4.5 | Pi | SpaceXAI | 38.9% |
| 13 | Claude Sonnet 5 | Claude Code | Anthropic | 37.3% |
| 14 | Claude Sonnet 5 | Pi | Anthropic | 33.9% |
| 15 | GPT-5.6 Sol | Pi | OpenAI | 33.8% |
| 16 | GPT-5.5 | Pi | OpenAI | 33.6% |
| 17 | GPT-5.6 Sol | Codex | OpenAI | 33.3% |
| 18 | GPT-5.5 | Codex | OpenAI | 31.5% |
| 19 | GPT-5.6 Terra | Pi | OpenAI | 31.0% |
| 20 | GPT-5.6 Terra | Codex | OpenAI | 29.7% |
| 21 | GPT-5.6 Luna | Pi | OpenAI | 19.4% |
| 22 | GPT-5.6 Luna | Codex | OpenAI | 18.2% |

## Sources and downloads

- [Paper PDF](/papers/txbench-ab.pdf)
- [Original publication](https://latch.bio/txbench-ab)
- [Repository](https://github.com/latchbio/txbench-antibody-discovery)
- [Interactive benchmark](https://benchmarks.bio/txbench-ab)
- [JSON](https://benchmarks.bio/benchmarks/txbench-ab/latest.json)
- [CSV](https://benchmarks.bio/benchmarks/txbench-ab/latest.csv)
- [V1 cross-benchmark leaderboard](https://benchmarks.bio/results/v1/)