News
A trillion-parameter open model from Meituan, Mistral's sharpened sovereignty pitch, a sober agent reality check from Qwen's former lead, Baidu's Unlimited OCR, and the quiet end of Mechanical Turk — plus Hollywood's Seedance hypocrisy.
News
Mistral rises as Europe's sovereign alternative, NVIDIA's agents write their own robot and chip code, Anthropic pushes Claude into science and drug discovery — while a 26,000-student study exposes AI's hidden learning costs. The July 4th digest.
News
Small open models keep beating the giants — Bridgewater's finance test, Mistral's Leanstral 1.5, and a diffusion ASR release. Plus Meta's agent delays, benchmarks that undersell AI, a CVE explosion, and Claude Code's tangled China problem.
News
Silicon and capital dominated July 2: Anthropic-Samsung chip talks, Microsoft's $2.5B Frontier unit, OpenAI's Washington equity gambit, and Nvidia bankrolling startups — balanced by open tooling from Alibaba and Anthropic, and agents that automate everything from turbines to dating.
News
Anthropic restores Fable and Mythos after an export-control ban — but hidden surveillance and stealth price hikes make the case for local weights. Plus NVIDIA's diffusion LLM, Google's TabFM, Meta's compute cloud, and Cloudflare forcing AI firms to pay publishers.
News
Anthropic's Claude Sonnet 5 and locally-run Claude Science lead the day, while DeepSeek and Meituan prove China can scale without Nvidia, agent-payment rails multiply, and a false-premise jailbreak exposes brittle safety guardrails.
News
Scarcity meets sprawl: a $590B memory chip buildout, Anthropic's land-grab across clouds and statehouses, fresh local-first agent tooling, and a Claude Code malware scare headline June 29.
News
Efficiency beat spectacle on June 28: Liquid AI and VibeThinker shrink reasoning into tiny models, Coinbase and 360 squeeze Western pricing, and Princeton's CEO-Bench plus Ford's engineer rehires expose what AI still can't do.
News
Open-source momentum from DeepSeek and Meta meets the tangled politics of Anthropic's model access, a benchmark-cheating scandal at OpenAI, fresh labor data, and market jitters — June 27's AI news, synthesized for builders.
News
OpenAI's GPT-5.6 arrives gated by per-customer government approval, custom silicon breaks Nvidia's grip, a startup swaps Claude for DeepSeek to survive, and two studies expose how shaky coding benchmarks really are.
News
Open-source OCR and coding models land under MIT, OpenAI unveils a custom Broadcom chip while the White House asks it to slow-roll GPT-5.6, the US–Europe chip war escalates, and agents move from demo to deployment across Google, Notion, and a $320M gaming-data bet.
News
OpenAI tapes out its first custom chip with Broadcom, DFlash promises 15x faster inference, GLM-5.2 undercuts Claude on price, and agents invade Slack, Figma, and marketing pipelines — while companies start rationing tokens.