Atlas

Benchmarks

← All benchmarks

PostTrainBench (pre-v1.1, H100)

Indexes

Historical PostTrainBench weighted aggregate across seven downstream benchmarks and four base models under the original one-H100, ten-hour setting. These observations precede the July 28, 2026 v1.1 contamination, external-API and model-identity checks; they are not rescored v1.1 or v1.2 results.

Top models (higher is better)

ModelScore
GLM-5.234.3
Opus 4.834.1
Loading Atlas data…