Thinking Machines Lab released Inkling, a 975-billion-parameter mixture-of-experts model that activates about 41 billion parameters per task and was trained on 45 trillion tokens of text, image, audio and video, reasoning natively across all four -- unlike the closed flagship models from OpenAI, Anthropic and Google, its weights are downloadable and modifiable.
why it mattersIt's a bet that open, natively multimodal weights beat one-size-fits-all proprietary models -- from a lab with the funding and pedigree to make that bet credible.
TechCrunch
Google's updated Deep Think reasoning mode is now available in the Gemini app for AI Ultra subscribers and, for the first time, via the Gemini API to a limited set of researchers, engineers and enterprises; early testers include a Rutgers mathematician who used it to catch a logical flaw in a paper that had passed human peer review.
why it mattersMoving Deep Think from a consumer feature to an API signals Google wants it used as a research tool, not just a chat mode.
Google
Chinese open-weight models accounted for 41% of Hugging Face downloads this spring, overtaking US models, and the six most-used models on OpenRouter are now all open-weight releases from Chinese firms including Tencent, Xiaomi, DeepSeek, MiniMax and Z.ai.
why it mattersThe competitive frontier that matters to most developers may now be open-weight usage share, not closed-model benchmark scores.
TechCrunch