Salesforce

A series of models that have two different thinking budgets: One for thinking, one for the solution phase. The models are trained with a budget-constrained rollout strategy using GRPO.

As evident by the o3 and o4-mini release: Tool-calling will be a super important feature for the next generation of models. This model, released under a non-commercial license, is a step forward in this direction with good benchmark scores on the main tool-calling benchmarks: BFCL and tau-bench.

Small multimodal model built on Phi 3 (BLIP style).