TxBench-Antibody-Development V1 (60 evaluations)

Science

Tests whether 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. 60 evaluations built from real experimental data, each graded deterministically. benchmarks.bio excludes it from its overall Capabilities score.

Benchmark source: Benchmarks

Top models (higher is better)

Model	Score	Source
Claude Sonnet 5.5	36.7	benchmarks.bio V1 published results · benchmarks.bio
GPT-6 Astra	32.8	benchmarks.bio V1 published results · benchmarks.bio
Gemini 4 Argon	31.1	benchmarks.bio V1 published results · benchmarks.bio
Grok 4.6	28.3	benchmarks.bio V1 published results · benchmarks.bio
GPT-6.1 Sol	27.8	benchmarks.bio V1 published results · benchmarks.bio
GPT-6.1 Terra	22.8	benchmarks.bio V1 published results · benchmarks.bio
GPT-5.6 Sol	18.9	benchmarks.bio V1 published results · benchmarks.bio
GPT-6 Luna	16.1	benchmarks.bio V1 published results · benchmarks.bio
Gemini 3.8 Flash	15.0	benchmarks.bio V1 published results · benchmarks.bio
DeepSeek V4.1 Flash	14.4	benchmarks.bio V1 published results · benchmarks.bio
Sonnet 5	13.3	benchmarks.bio V1 published results · benchmarks.bio
GLM-5.3	10.0	benchmarks.bio V1 published results · benchmarks.bio
Kimi K3	6.1	benchmarks.bio V1 published results · benchmarks.bio
Mistral Large 4	6.1	benchmarks.bio V1 published results · benchmarks.bio
GPT-5.6 Luna	5.0	benchmarks.bio V1 published results · benchmarks.bio
Nemotron 3.5 Lightning 30B A3B (precision unspecified)	4.4	benchmarks.bio V1 published results · benchmarks.bio