Interesting model release. It offers ~same capabilities as gpt-oss-120b but with faster inference (should be cheaper as well) and they're also providing (some?) pre-training and (all?) post-training datasets. In some benchmarks it is behind qwen3.5-122b that recently released. Focus on "agentic" stuff, and nvfp4 native. Glad to see nvda pushing for more models, more accessible models, and opening up some of their datasets.
Comments
Interesting model release. It offers ~same capabilities as gpt-oss-120b but with faster inference (should be cheaper as well) and they're also providing (some?) pre-training and (all?) post-training datasets. In some benchmarks it is behind qwen3.5-122b that recently released. Focus on "agentic" stuff, and nvfp4 native. Glad to see nvda pushing for more models, more accessible models, and opening up some of their datasets.