benchmarks.bio

Therapeutics · Capabilities

TxBench-Antibody-Discovery results

Agentic benchmark tasks for antibody discovery and development decisions.

Published score: Current V1 evaluation set (100 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for TxBench-Antibody-Discovery. Agentic benchmark tasks for antibody discovery and development decisions. Current V1 covers 100 evaluations and 45 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Claude Code, LSVP, 67.7%); Claude Opus 5.5* (Pi, LSVP, 66.7%); Gemini 4 Carbon (Pi, 65.0%). The highest-ranked published run without an LSVP exemption is Gemini 4 Carbon (Pi, 65.0%). Scores report benchmark-specific biological task performance on a 0–100 percentage scale. See the linked paper and repository for task construction, grading and uncertainty methodology. Results describe the complete model and agent harness used for each run, and are not comparable across different benchmarks or evaluation versions.

Published TxBench-Antibody-Discovery scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5*Claude CodeAnthropic67.7%LSVP
2Claude Opus 5.5*PiAnthropic66.7%LSVP
3Gemini 4 CarbonPiGoogle65.0%
4Gemini 4 ArgonPiGoogle63.7%
5Claude Fable 5.1*PiAnthropic63.3%LSVP
6Gemini 4 HydrogenPiGoogle57.7%
7Claude Sonnet 5.5Claude CodeAnthropic57.3%
8GPT-6 AstraPiOpenAI56.0%
9Claude Sonnet 5.5PiAnthropic54.7%
10Grok 4.6PiSpaceXAI53.0%
11GPT-6 AstraCodexOpenAI52.7%
12Grok 4.7PiSpaceXAI52.0%
13Grok 4.6Grok BuildSpaceXAI51.7%
14Gemini 3.7 FlashPiGoogle50.7%
15GPT-6.1 SolPiOpenAI49.7%
16Gemini 3.8 FlashPiGoogle49.0%
17Claude Opus 5PiAnthropic48.7%
18Claude Opus 5Claude CodeAnthropic48.0%
19GPT-6.1 SolCodexOpenAI45.7%
20GPT-6 SolPiOpenAI43.3%
21GPT-6.1 TerraPiOpenAI40.7%
22DeepSeek V4.1 FlashPiDeepSeek39.7%
23GPT-6.1 TerraCodexOpenAI38.7%
24GPT-5.6 SolPiOpenAI37.3%
25GPT-5.6 SolCodexOpenAI37.0%
26Claude Opus 4.7Claude CodeAnthropic36.0%
27Claude Opus 4.8Claude CodeAnthropic35.3%
28Kimi K3PiMoonshot35.3%
29Beam 1.0PiReflection34.0%
30GPT-6 LunaPiOpenAI34.0%
31Claude Opus 4.8PiAnthropic33.3%
32GPT-5.5PiOpenAI33.0%
33Claude Sonnet 5Claude CodeAnthropic32.0%
34Claude Opus 4.7PiAnthropic31.7%
35Mistral Large 4.0PiMistral28.3%
36Claude Sonnet 5PiAnthropic27.7%
37Claude Sonnet 4.6PiAnthropic26.3%
38GLM 5.3PiZ.ai26.3%
39Claude Sonnet 4.6Claude CodeAnthropic24.3%
40Nemotron 3 Ultra 550B A55BPiNVIDIA21.0%
41GPT-5.6 LunaPiOpenAI19.3%
42GPT-5.6 LunaCodexOpenAI18.7%
43Nemotron 3.5 LightningPiNVIDIA16.0%
44Claude Opus 5.5PiAnthropic8.3%
45Claude Opus 5.5Claude CodeAnthropic8.0%

Methodology and provenance

Scores use a 0–100 percentage scale. See Paper PDF, Original publication, Repository for task construction, grading, aggregation, and uncertainty details.

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.