# TxBench-Antibody-Development results > Tests whether AI agents can make the evidence-based decisions that take a therapeutic antibody from a chosen candidate to its first dose in people, such as whether it can be made and stored reliably, whether a manufacturing change altered the drug, or what starting dose is safe. Data updated: 2026-10-30 Artifact generated: 2026-11-04T17:19:37.999Z - Domain: Therapeutics - Horizon: short - Assessment: Capabilities - Published score scope: Current V1 evaluation set (60 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. 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. | Rank | Model | Harness | Provider | Score | Variant | | ---: | --- | --- | --- | ---: | --- | | 1 | Grok 4.7 | Grok Build | SpaceXAI | 39.4% | | | 2 | Claude Opus 5* | Claude Code | Anthropic | 37.2% | LSVP | | 3 | Claude Sonnet 5.5 | Pi | Anthropic | 36.7% | | | 4 | Grok 4.7 | Pi | SpaceXAI | 36.1% | | | 5 | Claude Sonnet 5.5 | Claude Code | Anthropic | 34.4% | | | 6 | Claude Opus 5.5* | Pi | Anthropic | 32.8% | LSVP | | 7 | Claude Opus 5.5* | Claude Code | Anthropic | 32.8% | LSVP | | 8 | GPT-6 Astra | Pi | OpenAI | 32.8% | | | 9 | Gemini 4 Argon | Pi | Google | 31.1% | | | 10 | GPT-6 Astra | Codex | OpenAI | 31.1% | | | 11 | Claude Opus 5* | Pi | Anthropic | 30.0% | LSVP | | 12 | Grok 4.6 | Grok Build | SpaceXAI | 28.3% | | | 13 | Claude Fable 5.1 | Pi | Anthropic | 27.8% | | | 14 | GPT-6.1 Sol | Codex | OpenAI | 27.8% | | | 15 | Claude Fable 5.1 | Claude Code | Anthropic | 26.9% | | | 16 | GPT-6.1 Sol | Pi | OpenAI | 26.7% | | | 17 | Grok 4.6 | Pi | SpaceXAI | 26.1% | | | 18 | GPT-6.1 Terra | Codex | OpenAI | 22.8% | | | 19 | GPT-6.1 Terra | Pi | OpenAI | 22.2% | | | 20 | GPT-5.6 Sol | Pi | OpenAI | 18.9% | | | 21 | GPT-6 Luna | Codex | OpenAI | 18.9% | | | 22 | GPT-6 Luna | Pi | OpenAI | 16.1% | | | 23 | Gemini 3.8 Flash | Pi | Google | 15.0% | | | 24 | DeepSeek V4.1 Flash | Pi | DeepSeek | 14.4% | | | 25 | Claude Sonnet 5 | Pi | Anthropic | 13.3% | | | 26 | GPT-5.6 Sol | Codex | OpenAI | 12.2% | | | 27 | Claude Sonnet 5 | Claude Code | Anthropic | 11.1% | | | 28 | GLM 5.3 | Pi | Z.ai | 10.0% | | | 29 | Kimi K3 | Pi | Moonshot | 6.1% | | | 30 | Mistral Large 4.0 | Pi | Mistral | 6.1% | | | 31 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 5.6% | | | 32 | GPT-5.6 Luna | Pi | OpenAI | 5.0% | | | 33 | Nemotron 3.5 Lightning | Pi | NVIDIA | 4.4% | | | 34 | GPT-5.6 Luna | Codex | OpenAI | 2.8% | | ## Sources and downloads - [Paper PDF](/papers/txbench-ab-dev.pdf) - [Repository](https://github.com/latchbio/txBench-antibody-development) - [Interactive benchmark](https://benchmarks.bio/txbench-abdev) - [JSON](https://benchmarks.bio/benchmarks/txbench-abdev/latest.json) - [CSV](https://benchmarks.bio/benchmarks/txbench-abdev/latest.csv) - [Current V1 cross-benchmark leaderboard](https://benchmarks.bio/results/v1/)