Models, machine learning, applied agents, and the fast-moving world of AI.
19 articlesMeta shipped its biggest open-weight model this week, and a distilled 30B agent that plans, calls tools, and recovers from errors on a single consumer GPU.
Anthropic's three-week Claude 5.5 refresh cut flagship prices and gave the $0.10 Haiku real agentic chops.
The 2026 multi-agent framework choice is less about vendor and more about control — role-based crews to prototype, graph-based engines to ship.
DeepMind's AlphaGo beat Go world champion Lee Sedol 4-1 in Seoul, March 2016 — a win built on search, neural nets, and one unforgettable move.
Near-Pro reasoning from just 13B activated parameters. DeepSeek's fully open V4-Flash MoE is MIT-licensed, self-hostable, and cheap to serve.
Xiaomi's MIT-licensed MiMo-V2.6-Pro, a 1.02T sparse-MoE model, tops the open-weight field at 46.32 on the Artificial Analysis Intelligence Index.
