Compare current model-and-harness results for TxBench-Preclinical-Pharmacology. Agentic benchmark tasks for preclinical pharmacology analysis and decision-making. Current V1 covers 100 evaluations and 44 ranked configurations. Data updated 2026-11-13. The highest-ranked published configurations are Gemini 4 Carbon (Pi, 66.0%); Claude Opus 5.5* (Pi, LSVP, 65.3%); Claude Opus 5.5* (Claude Code, LSVP, 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.
| Rank | Model | Harness | Provider | Score | Variant |
|---|---|---|---|---|---|
| 1 | Gemini 4 Carbon | Pi | 66.0% | ||
| 2 | Claude Opus 5.5* | Pi | Anthropic | 65.3% | LSVP |
| 3 | Claude Opus 5.5* | Claude Code | Anthropic | 65.0% | LSVP |
| 4 | Gemini 4 Argon | Pi | 64.7% | ||
| 5 | Claude Sonnet 5.5 | Pi | Anthropic | 64.3% | |
| 6 | Claude Opus 5 | Pi | Anthropic | 63.0% | |
| 7 | GPT-6 Sol | Pi | OpenAI | 63.0% | |
| 8 | Claude Sonnet 5.5 | Claude Code | Anthropic | 62.3% | |
| 9 | Grok 4.6 | Pi | SpaceXAI | 61.7% | |
| 10 | Gemini 4 Hydrogen | Pi | 61.3% | ||
| 11 | GPT-5.6 Sol | Pi | OpenAI | 61.3% | |
| 12 | Claude Fable 5.1* | Pi | Anthropic | 59.7% | LSVP |
| 13 | Claude Opus 5 | Claude Code | Anthropic | 59.7% | |
| 14 | GPT-6.1 Sol | Pi | OpenAI | 59.7% | |
| 15 | GPT-6 Astra | Pi | OpenAI | 59.3% | |
| 16 | GPT-6 Astra | Codex | OpenAI | 59.3% | |
| 17 | Claude Opus 4.8 | Pi | Anthropic | 57.7% | |
| 18 | Grok 4.6 | Grok Build | SpaceXAI | 57.0% | |
| 19 | GPT-6.1 Terra | Pi | OpenAI | 55.7% | |
| 20 | Gemini 3.7 Flash | Pi | 55.3% | ||
| 21 | GPT-5.5 | Pi | OpenAI | 55.3% | |
| 22 | GPT-6.1 Sol | Codex | OpenAI | 55.0% | |
| 23 | Gemini 3.8 Flash | Pi | 54.7% | ||
| 24 | Grok 4.7 | Pi | SpaceXAI | 54.7% | |
| 25 | Claude Opus 4.8 | Claude Code | Anthropic | 54.3% | |
| 26 | GPT-6.1 Terra | Codex | OpenAI | 50.7% | |
| 27 | Claude Sonnet 5 | Pi | Anthropic | 49.3% | |
| 28 | Claude Sonnet 5 | Claude Code | Anthropic | 49.3% | |
| 29 | DeepSeek V4.1 Flash | Pi | DeepSeek | 46.7% | |
| 30 | Claude Opus 4.7 | Pi | Anthropic | 45.3% | |
| 31 | GPT-6 Luna | Pi | OpenAI | 44.3% | |
| 32 | GPT-5.6 Luna | Codex | OpenAI | 43.7% | |
| 33 | GPT-5.6 Luna | Pi | OpenAI | 43.3% | |
| 34 | Claude Opus 4.7 | Claude Code | Anthropic | 42.3% | |
| 35 | Mistral Large 4.0 | Pi | Mistral | 42.3% | |
| 36 | Beam 1.0 | Pi | Reflection | 39.7% | |
| 37 | Kimi K3 | Pi | Moonshot | 39.7% | |
| 38 | GLM 5.3 | Pi | Z.ai | 35.0% | |
| 39 | Claude Sonnet 4.6 | Pi | Anthropic | 34.3% | |
| 40 | Claude Sonnet 4.6 | Claude Code | Anthropic | 33.3% | |
| 41 | Claude Opus 5.5 | Claude Code | Anthropic | 30.7% | |
| 42 | Claude Opus 5.5 | Pi | Anthropic | 26.3% | |
| 43 | Nemotron 3 Ultra 550B A55B | Pi | NVIDIA | 22.7% | |
| 44 | Nemotron 3.5 Lightning | Pi | NVIDIA | 10.3% |
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.