A 70B parameter open-weight release topped several reasoning benchmarks this week.

Inference costs drop 3x versus comparable closed models.

Commercial licensing terms allow enterprise deployment.