Atlas

Benchmarks

← All benchmarks

FrontierSWE: PCQM4Mv2 Molecular Gap Prediction (Mean@5)

Code · 2026-04-16

Train a 2D-only molecular graph regressor for a PCQM4Mv2-derived task. FrontierSWE Mean@5 is the average normalized task score across up to five independent long-horizon agent runs.

Top models (higher is better)

ModelScore
GLM-5.20.9
Claude Fable 50.9
Grok 4.50.9
GPT-5.40.8
Composer 2.50.8
Opus 4.60.5
GPT-5.50.5
Opus 4.70.4
Opus 4.80.4
GLM-5.10.2
DeepSeek-V4-Pro0.2
Gemini 3.1 Pro Preview0.2
Qwen3.6 Plus (2026-04-02)0.2
Kimi K2.50.2
Kimi K2.60.0
Loading Atlas data…