Multi-token Residual Prediction (MRP) is a lightweight module for diffusion language models that predicts the residual between adjacent denoising steps rather than the full distribution. This enables both lossless speedup in static regimes (up to 1.56×) and quality recovery in dynamic regimes (up to +16 accuracy points) without trade-offs.

Hugging Face and Cerebras have collaborated to create a real-time voice AI system powered by Gemma 4, achieving dramatically lower latency through an open modular architecture. The pipeline integrates Nvidia's speech recognition, Cerebras's fast inference, and Alibaba's text-to-speech, and is already deployed in over 9,000 Reachy Mini robots.

Together AI raised $800M in Series C funding to accelerate the shift to open-source AI. The company argues that closed models' economics don't scale, and open models combined with full-stack optimization can achieve 6-20x cost reductions. Together AI has launched innovations like FlashAttention-4 and Together Megakernel, becoming one of the world's largest AI token producers.

Anthropic announces that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, with access restoration starting tomorrow.

Google releases Nano Banana 2 Lite (aka Gemini 3.1 Flash Lite Image), billed as the fastest and cheapest Gemini image model. The author tested it with a 'Where's Waldo' prompt and got good results despite spelling errors.

Anthropic releases Claude Sonnet 5, with performance close to Opus 4.8 at lower prices. The model introduces a new tokenizer that increases token count by approximately 30% for the same input, effectively raising costs. API changes include removal of sampling parameters and adaptive thinking enabled by default.