Process reward and policy models to enable parallel test time inference and scoring of multiple traces by a generator model. This is (likely) similar to the implementation of Grok Heavy, Gemini Deep Think and GPT-5 pro.
Specs
Params32B
LicenseApache-2.0
Tags
modelsmodels/llmsmodels/llms/reasoningmultilingualartifacts/13
Related Models
