← All benchmarks

STEMBench120 v1.0.0 (with tools)

Science

The same 120 STEMBench120 v1.0.0 research tasks and complete-run mean fractional progress score (0-4), run as tool-enabled agent configurations: Codex with web or source retrieval, code execution, file operations and multi-day agent workflows. Ulam presents these as complete systems under different resource settings from the closed-book API runs, so they are kept apart from that benchmark and outside the capability fit.

Top models (higher is better)