Overview
- Kimi announced K3 after a July 17 launch as an open model with 2.8 trillion parameters, native visual understanding and a 1,000,000‑token context window.
- K3 topped the Frontend Code Arena leaderboard with a 1679 score on front‑end code tasks, though Kimi says the model still trails the strongest closed models such as Claude Fable 5 and GPT‑5.6 Sol on overall performance.
- The company confirmed it will publish full model weights by July 27, a move that would let developers run and inspect the model directly and speed community research and tooling.
- Kimi reported a record single‑day rise in annualized revenue after the release and temporarily paused new consumer subscriptions to focus existing compute on current users while it expands capacity.
- Coverage from Bloomberg and others says the release has prompted investors and engineers to reassess the US‑China lead in cutting‑edge AI, and separate media reports that Kimi is preparing a Hong Kong IPO remain unconfirmed and developing.