[AINews] It's Meta-Harness Summer
OpenAI announced Jalapeño, its first custom AI chip for LLM inference, built with Broadcom and intended for ChatGPT, Codex, and future agent products, signaling a push to own more of the AI stack. Mea…
OpenAI announced Jalapeño, its first custom AI chip for LLM inference, built with Broadcom and intended for ChatGPT, Codex, and future agent products, signaling a push to own more of the AI stack. Mea…
Databricks co-founders Matei Zaharia and Reynold Xin argue that the frontier AI ecosystem must remain open, unveiling Omnigent, an open-source meta-harness for combining and controlling agents across …
Anthropic launched Claude Tag, a Slack-native agent that teams can tag to delegate tasks asynchronously. The feature, which has been used internally to write 65% of the product team's code, marks a sh…
SpaceX has become a $28 billion per year neocloud business through GPU rental deals with AI companies including Anthropic, Google, and Reflection AI, according to analyst Jamin Ball. The company's ann…
Gray Swan co-founders Zico Kolter and Matt Fredrikson discussed the state of AI red teaming, warning that prompt injection and agent vulnerabilities create a new class of security risks that tradition…
Latent.Space is offering a $250 discount on AI Engineer conference tickets to its paying subscribers until Monday. The promotion was announced in AINews, but many subscribers have not yet opted in to …
Zhipu AI's GLM-5.2 open-weight model has passed multiple real-world 'vibe checks,' with Jeremy Howard and Artificial Analysis rating it comparable to GPT 5.5 and Opus 4.8, marking a shift from benchma…
Anjney Midha, founder of AMP, argues that the AI scaling race is not just about acquiring more GPUs but about maximizing the efficiency of existing hardware, citing xAI's sub-10% Model FLOPs Utilizati…
Midjourney announced a full-body ultrasonic CT scanner and a spa venture in San Francisco, aiming to revolutionize medical imaging with a device that uses 358,000 ultrasonic elements and claims 0.5mm …
Radical AI founder Joseph Krause says the company's self-driving lab produced and characterized 1,200 alloys in six months, a nearly 10x speedup over the DARPA/GE MACH program. The lab uses an AI scie…
Z.ai released GLM-5.2, an MIT-licensed open-weight frontier model with 744B parameters, achieving top scores in frontend coding benchmarks and surpassing Opus 4.8. The model features a 1M-token contex…
OpenAI's 'Practical Guide to Building Agents' has sparked a debate with LangChain CEO Harrison Chase, who called it 'misguided' and published a detailed rebuttal. The conflict highlights a core tensio…
Anthropic suspended access to its Claude Fable 5 and Mythos 5 models for all customers after a US government directive citing a potential jailbreak as a national cybersecurity risk, a move Anthropic d…
Anthropic rolled out its Claude Fable 5 model with a covert policy that silently degraded performance for AI-research-related use cases, then reversed the policy within roughly a day after public back…
Sarah Guo, a prominent AI investor and Cognition backer, published an essay arguing that the most defensible AI companies will be those focused on "untrainable" work—integrating models into messy, pri…
Anthropic launched Claude Fable 5, a Mythos-class model at least twice the size of Opus 4.8, with API pricing at roughly 2x Opus and benchmark improvements including a jump from 13.4% to 29.3% on the …
Cognition introduced FrontierCode, a new benchmark that evaluates code on mergeability rather than just unit-test passing, with tasks built by open-source maintainers requiring over 40 hours each. The…
Auriel W, an RL practitioner who has worked at Gemini, warns that low-quality reinforcement learning environments with broken code, race conditions, and caching bugs are systematically ruining model t…
Anthropic published research on recursive self-improvement (RSI) while OpenAI's ChatGPT crossed 1 billion monthly active users five months behind schedule and announced improved memory features. NVIDI…
Andon Labs co-founders Lukas Petersson and Axel Backlund are stress-testing frontier AI models by giving them real-world control over inventory, wallets, and physical storefronts, revealing unexpected…