I think multi-agent systems are being oversold as a substitute for a stronger model. AutoFyn improved Sonnet, Opus and GLM on IMO 2026, but every run missed the same decisive reduction. More agents can search harder. They cannot invent an idea the base model never reaches.
Microsoft is reportedly testing Kimi K3 for Copilot. Four years ago the model name was the product. Now Copilot may keep the same interface while swapping OpenAI, Anthropic or Moonshot underneath. The model is starting to look more like a cloud component than a brand
Sanctioning Moonshot would not delete Kimi weights already downloaded. It would stop U.S. clouds and companies from serving them.
its less about erasing open model and more about chosing which businesses are allowed to make it convenient
Claude Opus 5: Launching Today
- Claude Opus 5 is scheduled to launch today (Thursday for most time zones), assuming there are no last-minute delays.
- As I mentioned earlier, Anthropic has been targeting a Thursday release.
- Benchmark performance will be comparable to Fable 5, but won't surpass it.
Claude Opus 5: Launching Today
- Claude Opus 5 is scheduled to launch today (Thursday for most time zones), assuming there are no last-minute delays.
- As I mentioned earlier, Anthropic has been targeting a Thursday release.
- Benchmark performance will be comparable to Fable 5, but won't surpass it.
I would rather watch Send Help take a big tonal swing and become slightly ridiculous than watch another perfectly polished thriller with no personality.
Even when Sam Raimi overdoes it, the movie still feels directed by an actual person.