Atlas

Benchmarks

← All benchmarks

PostTrainBench Lite - release-reported performance

Code

Named Lite variant of PostTrainBench in OpenAI release tables. Task set, hardware and submission budget are not identified; diagnostic until equivalence to a versioned public protocol is established.

Top models (higher is better)

ModelScore
GPT-5.6 Terra51.5
GPT-5.6 Sol50.3
GPT-5.6 Luna29.6
Loading Atlas data…