Atlas

Benchmarks

← All benchmarks

Terminal-Bench-Science 0.1: 6× Verifier Timeout

Agents

Google's self-computed Terminal-Bench-Science 0.1 result with verifier timeouts multiplied by six. This verification protocol differs from the standard leaderboard and is retained as a diagnostic rather than pooled with standard-timeout scores.

Top models (higher is better)

ModelScore
Gemini 4 Argon57.6
Loading Atlas data…