a bit less capable but ~comparable to qwen 3.5 122b ~ 2x faster inference than qwen 3.5 122b ~ 7x faster inference than gpt-oss 120b probably most important:

a bit less capable but ~comparable to qwen 3.5 122b ~ 2x faster inference than qwen 3.5 122b ~ 7x faster inference than gpt-oss 120b probably most important: training datasets and training recipe available (!) in other words this is an open source llm release (not just open weights!)