Atlas

Benchmarks

← All benchmarks

FrontierSWE: PCQM4Mv2 Molecular Gap Prediction (Best@5)

Code · 2026-04-16

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

Top models (higher is better)

ModelScore
Claude Fable 50.9
Opus 4.60.9
Opus 4.80.9
GLM-5.20.9
Grok 4.50.9
Opus 4.70.9
GPT-5.50.9
Composer 2.50.9
GLM-5.10.9
GPT-5.40.9
DeepSeek-V4-Pro0.9
Qwen3.6 Plus (2026-04-02)0.8
Gemini 3.1 Pro Preview0.8
Kimi K2.50.8
Kimi K2.60.0
Loading Atlas data…