According to Thinking Machines Lab: Inkling 2 Small is a new Mixture-of-Experts AI model with 594B total parameters, offering massive upgrades in coding, vision, and reasoning. Initially available with closed weights on Tinker, its full Apache 2.0 weights will be released next week following safety checks.
According to Apple Machine Learning: Researchers at Apple have developed a new method to speed up the core search process behind recommendation systems and RAG. Instead of treating all search queries the same, they use small neural networks trained on typical user questions to predict and route queries to the right data much faster. This approach significantly cuts down search latency while keeping high accuracy, offering a smarter way to handle massive databases.
According to Mistral AI News: Mistral has updated its enterprise platform, Mistral Studio, alongside its new AI model, Mistral Large 4. The update gives businesses the tools to build, run, and manage AI agents, workflows, and applications securely. Companies can deploy these tools in the cloud or on their own private servers while keeping full control over data.
According to World Labs: Researchers have released Frameboy, a new demo that lets users explore 3D worlds generated by the Atlas world model in real-time at up to 15 fps. By streaming frames from cloud GPUs straight to the browser, the tool removes the frustrating wait times of traditional AI video generation. This speed lets creators explore and iterate instantly, completely changing how people interact with generative 3D spaces.
According to Cerebras Blog: Cerebras CEO Andrew Feldman published a detailed response to a short seller report claiming that their massive wafer-scale AI chips wear out in two years. Feldman explained that they solved the extreme thermal expansion challenges years ago through steady temperature controls and flexible electrical connections.…
SplatAnything introduces Gaussian Spacing Normalization (GSNorm) to canonicalize scene-level Gaussian spacing, resolving ambiguous supervision and improving feed-forward 3D Gaussian Splatting across varying scene scales and viewpoints.
According to Oumi: Oumi has updated its automated AI engineer called FAiE to support building and testing complete AI agents rather than just single models. The new release also brings cost tracking and advanced training methods.
According to LlamaIndex Blog: Document AI tool LlamaParse can now be deployed wherever your data lives, ranging from a fully managed SaaS to private VPCs and air-gapped data centers. Teams can choose whether to manage their own GPUs or let LlamaIndex handle inference, matching strict security and compliance needs.
