Mistral Large 4 ("Le Chonk") is a 1T-parameter natively multimodal model with 49B active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's European datacenters. The preview API is live on Mistral Studio; weights drop end of this month.
Surface Laptop Ultra starts at $2,599 with Nvidia's RTX Spark SoC, up to 128GB unified memory, shipping October 16; the Surface RTX Spark Dev Box costs $5,999 and claims one petaflop of AI compute. Microsoft announced general availability of Microsoft Execution Containers (MXC) for running agents under OS control, plus Copilot "Hybrid Intelligence" with local file access.
Amazon is negotiating to acquire AI startup Decart AI in a deal that could be valued at around $7 billion, per the Wall Street Journal. The report cites unnamed sources and no terms have been confirmed.
EmbeddingGemma 2 maps text, code, images, video, and audio into one 768-dimensional space under Apache 2.0, built on Gemma 4. It scales from 270M parameters for text-only to 740M for all modalities, with Matryoshka truncation cutting vectors to 128 dims for up to 6x storage savings.
Step 5 Preview is a sparse MoE with 600B total and 27B active parameters, a 1M-token context window, and 64K max output, priced at $1/M input tokens. It targets agentic coding, professional knowledge work, and finance, and is available via StepFun AI Studio, APIs, and Vercel AI Gateway.
Robinson, who led safety reports for OpenAI's product launches and spent three-and-a-half years there, resigned in an Atlantic essay titled "I quit OpenAI because its culture is broken." He says frontier labs need nuclear-plant-style redundancy, warning "iteration after a mistake may not be possible."