Model
- DeepSeek ships V4.1-Flash-1017: higher coding and agent scores, shorter outputs, MIT weights and a Search API
- StepFun Step 5 Preview: announced Sept 20, weights visible on the Oct 15 plan
- Meta opens Muse Spark Open weights: an 816B multimodal MoE with 1M context, licensed with a separate deal for big model-service providers
- Microsoft AI finishes MAI-Base-2 pre-training: 2.8T MoE, claims up to ~12× efficiency vs DeepSeek base
- Grok Voice 3.0 reaches the Voice Agent API at the same price, with an automatic latest-alias switch
- Liquid AI releases LFM2.5-24B-A2B, an open-weight model for local agents
- Reflection announces Beam, a 501-billion-parameter open-weight model; weights are still pending
- GPT-6 Sol and Luna launch with lower API prices and improved caching
- Qwen-Image-2.1 combines generation and transparent editing; commercial use needs authorization
- Gemini 3.8 Live lets voice assistants keep talking while tools run