The Syndicate

AI news.No slop.Just facts.

What actually happened in AI today. Short, daily, free.

Tuesday 27 October

Open-weight coding model tops the agent benchmark: Local coding agents stop being a compromise.
Three labs cut inference prices in the same week: Running a daily agent gets cheaper again.
Browser agents get a shared protocol for logins: Fewer agents stuck at a sign-in screen.
EU drafts rules for agents that act on your behalf: Agents that buy or book may need a clear human sign-off.
A 3B model matches last year’s 70B on reasoning tests: Small models keep getting good enough for phones.
Voice assistants learn to wait their turn: Talking over people was the top complaint about voice agents.
A startup raises $120M for agent payment rails: Agents that buy things need a safe way to pay.
Open model hub passes 2 million public models: Most are small fine-tunes you can run locally.