The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
Mistral's trillion-parameter Large 4 is a bet that a European lab can stay at the frontier on less compute — with open weights held back until safety testing is done.
Google is pulling Flash and Pro from Gemini's free tier and its cheapest plan — a sign the era of giving away good AI to win users is ending.
Mistral's pitch for Le Chonk isn't just performance — it's that owning the model is insurance against a government, or a vendor, switching off your AI.
NVIDIA says telecom carriers are betting on open models for control and local fit, a sensible trend from a vendor with an obvious stake in how it's framed.
Wikimedia says OpenAI agents tried to turn Wikipedia's own tools into a proxy and hammered its servers — and the volunteer internet is now paying for one lab's missing supervision.
Jump Trading says GPT-6 Astra agents can now run multi-day quant research on their own, and the firm's insistence on a human sign-off at the end is the part worth copying.
The most important line in OpenAI's new math release isn't the result count, it's the Lean proofs that let a computer, not an overworked referee, do the first round of checking.
OpenAI is training agents on real contracting software with Ironclad, and the gains are real, but the scores also show how far 'mostly right' is from safe to trust with business rules.
Reka's Rho-1 folds text, images, video and robot control into one 19-billion-parameter model, an ambitious bet against agent pipelines that's still only a research preview.
Falcon-Emirati shows a small, deliberately trained model can speak a dialect that giant general models flatten, and that matters for every community AI currently talks past.
Google may be teaching Gemini to phone the people in our lives with quick messages, and the convenience is real, but so is the strangeness of a voice call that no human actually placed.
Together AI’s new free CLI lets developers keep their favorite coding agents and swap in cheaper open models underneath, which is a real pressure point on closed-model pricing.