benchmarks.bio

Therapeutics · Capabilities

TxBench-Oligo-Discovery results

Agentic benchmark tasks for oligonucleotide discovery and development decisions.

Published score: Current V1 evaluation set (113 evaluations) · short horizon · data updated 2026-11-13 (November 13, 2026)

Compare current model-and-harness results for TxBench-Oligo-Discovery. Agentic benchmark tasks for oligonucleotide discovery and development decisions. Current V1 covers 113 evaluations and 45 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Claude Opus 5.5* (Claude Code, LSVP, 62.2%); Claude Opus 5.5* (Pi, LSVP, 59.9%); Gemini 4 Carbon (Pi, 55.8%). The highest-ranked published run without an LSVP exemption is Gemini 4 Carbon (Pi, 55.8%). 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-Oligo-Discovery scores
RankModelHarnessProviderScoreVariant
1Claude Opus 5.5*Claude CodeAnthropic62.2%LSVP
2Claude Opus 5.5*PiAnthropic59.9%LSVP
3Gemini 4 CarbonPiGoogle55.8%
4Claude Fable 5.1*PiAnthropic54.9%LSVP
5Gemini 4 HydrogenPiGoogle54.3%
6GPT-6.1 SolPiOpenAI52.5%
7Claude Sonnet 5.5Claude CodeAnthropic51.9%
8Claude Sonnet 5.5PiAnthropic51.9%
9GPT-6 AstraPiOpenAI51.9%
10Gemini 4 ArgonPiGoogle51.0%
11GPT-6 AstraCodexOpenAI50.7%
12GPT-6.1 SolCodexOpenAI49.6%
13Claude Opus 5Claude CodeAnthropic48.4%
14Grok 4.6PiSpaceXAI46.0%
15Grok 4.6Grok BuildSpaceXAI44.3%
16GPT-6 SolPiOpenAI43.1%
17Claude Opus 5PiAnthropic42.5%
18Grok 4.7PiSpaceXAI41.9%
19Gemini 3.7 FlashPiGoogle37.8%
20Claude Opus 4.8PiAnthropic37.5%
21GPT-5.6 SolCodexOpenAI37.2%
22Claude Sonnet 5PiAnthropic36.6%
23Claude Opus 4.8Claude CodeAnthropic36.0%
24GPT-5.6 SolPiOpenAI36.0%
25Gemini 3.8 FlashPiGoogle35.1%
26DeepSeek V4.1 FlashPiDeepSeek34.8%
27Claude Sonnet 5Claude CodeAnthropic33.9%
28GPT-5.5PiOpenAI33.9%
29GPT-6 LunaPiOpenAI33.6%
30Claude Opus 4.7PiAnthropic31.9%
31GLM 5.3PiZ.ai31.9%
32Claude Opus 4.7Claude CodeAnthropic31.3%
33Kimi K3PiMoonshot31.3%
34GPT-6.1 TerraPiOpenAI30.4%
35Claude Sonnet 4.6PiAnthropic30.1%
36GPT-6.1 TerraCodexOpenAI28.3%
37Claude Sonnet 4.6Claude CodeAnthropic28.0%
38Beam 1.0PiReflection27.4%
39Mistral Large 4.0PiMistral27.1%
40Nemotron 3 Ultra 550B A55BPiNVIDIA25.1%
41GPT-5.6 LunaCodexOpenAI23.0%
42GPT-5.6 LunaPiOpenAI23.0%
43Nemotron 3.5 LightningPiNVIDIA8.6%
44Claude Opus 5.5PiAnthropic1.5%
45Claude Opus 5.5Claude CodeAnthropic1.2%

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.