Microsoft launchedseven in-house MAI models, headlined by MAI-Thinking-1,a 35B-active MoE trained from scratch with no distillation. Sources: MAI keynote transcript, CNBC, GitHub changelog
StepFun Step 3.7 Flash: an open ~198B MoE vision-language model (~11B active, Apache 2.0) built for production agents, up to 400 tok/s. Sources: StepFun blog, Hugging Face
Grok V9-Medium finished pretraining at 1.5T parameters (3x the production V8-Small), per Musk, with release targeted mid-June, notably trained on real Cursor developer-workflow data. Sources: TechTimes
Grok Imagine Video 1.5: image-to-video at 720p with natively generated synchronized audio, debuting #1 on the Image-to-Video Arena at $0.08-0.14/second. Sources: xAI, The Decoder
Comments
Sign in as a member to join the conversation.
Loading comments…