List your company

Scientific Coding on ScienceAgentBench deterministic-eval 2024

47.4Pass RateDeep negative knowledge retry
30.96835.23439.543.766Jun 19, 2026

Evaluation results

MethodLinks
Deep negative knowledge retryBackbone=Claude Sonnet 4.6, Number of feedback rounds=32026.0647.4795
Self-debugBackbone=Claude Sonnet 4.6, Number of feedback rounds=32026.0644.71,109
Negative knowledge retryBackbone=Claude Sonnet 4.6, Number of feedback rounds=12026.0636.8296
BaseBackbone=Claude Sonnet 4.6, Protocol=First attempt2026.0631.6—
RetryBackbone=Claude Sonnet 4.6, Protocol=Fresh agent, no memory2026.0631.6—