Musk v. OpenAI trial reveals xAI distilled from OpenAI models; Microsoft and OpenAI renegotiate partnership to allow multi-cloud deployment; DeepSeek releases V4 preview models with 1 million-token context; Google DeepMind introduces Vision Banana, unifying image generation and visual understanding.

Elon Musk testified over three days in Oakland that xAI 'partly' used OpenAI's models to train its own through distillation, though he characterized it as 'standard practice.' Musk is seeking up to $134 billion in damages, removal of Sam Altman and Greg Brockman, and an unwinding of OpenAI's for-profit conversion, citing that he co-founded OpenAI in 2015 as a nonprofit and donated approximately $38 million.

Microsoft and OpenAI replaced Microsoft's open-ended exclusivity clause with a nonexclusive license through 2032. OpenAI can now serve all products across any cloud provider including AWS, though OpenAI products ship 'first on Azure' unless Microsoft cannot support necessary capabilities. Microsoft stops paying OpenAI a revenue share, while OpenAI continues paying Microsoft a revenue share through 2030 (subject to a cap).

DeepSeek launched preview versions of V4 Flash (284 billion total parameters, 13 billion active) and V4 Pro (1.6 trillion total parameters, 49 billion active), both text-only mixture-of-experts models with 1 million-token context windows. Weights are open-sourced on Hugging Face with detailed technical reports; V4-Pro-Max is claimed to be almost uniformly better than recent open-source releases from China (Kimi-K.26 and GLM-5.1).

Google DeepMind introduced Vision Banana, a unified model built by instruction-tuning their base image generator Nano Banana Pro that performs both image generation and visual understanding tasks including semantic segmentation, instance segmentation, monocular metric depth estimation, and surface normal estimation. Vision Banana achieves a 53.5% win rate against Nano Banana Pro on GenAI-Bench text-to-image generation without degrading generative performance.