Moonshot AI has released Kimi K3 — and this is no longer "just another Chinese model"

Moonshot AI has released Kimi K3 — and this is no longer "just another Chinese model" Kimi has a new flagship: Kimi K3. According to official data — around 2.8 trillion parameters, context up to 1 million tokens, native image processing, and a focus on agents and code. What this means in practice: • You can throw in a large repository or a long document without losing the thread; • The model is optimized for long sessions: coding, refactoring, analyzing an entire project; • Thinking mode is essentially always on. Along with the base model, these modes were released: • K3 Max — chat + all-purpose agent; • K3 Swarm — a swarm of agents working in parallel on a single task (mass search, batches, complex pipelines). Early comparisons place K3 somewhere above Claude Opus 4.8, but still below Fable 5 — meaning it's already in the top league, not a "budget alternative." Another interesting point: Moonshot hints at open weights, promising to release them in the coming days. You can try it here: https://www.kimi.com/code