Kimi K3 is the first open model to crack 2.8T params
Moonshot AI's Kimi K3 becomes the largest open-weight model ever, tripling its predecessor and topping Claude Opus 4.8 on coding.
China's Moonshot AI just dropped Kimi K3 — the first open model in the 3-trillion-parameter class, landing at 2.8T parameters, nearly triple the size of Kimi K2.6. It's now the largest open-weight AI model on the planet.
Under the hood: a new hybrid linear attention scheme called Kimi Delta Attention (KDA), Attention Residuals, native visual understanding, and a 1M-token context window. On Moonshot's own evals it beat every open rival — including Claude Opus 4.8 and GPT 5.5 — across coding and agentic benchmarks, and even edged out Claude Fable 5 in the Frontend Code Arena. It still trails Fable 5 and GPT 5.6 Sol on overall performance.
Weights ship publicly by July 27 under a Modified MIT license — the latest jolt in China's open-weight push around US compute limits, and it's already reigniting regulation talk.
Why it matters: frontier-scale capability is now something you can download, not just rent from an API.
Sources
- Moonshot releases 2.8-trillion-parameter Kimi K3 tomshardware.com
- Chinese AI levels up with the open-weight shift cnbc.com
Written by an AI pipeline from the sources above. How it works.
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.