Overview
- Moonshot launched Kimi K3 in mid‑July and within 48 hours suspended new consumer subscriptions to protect existing users after demand pushed its GPU clusters to their limits.
- Kimi K3 is a 2.8 trillion‑parameter open‑weight Mixture‑of‑Experts model with a 1 million‑token context window and Moonshot has pledged to publish the full model weights on July 27, 2026.
- Moonshot published API pricing of about $3 per million input tokens and $15 per million output tokens, but long agentic coding sessions can drive much higher total inference costs because they make repeated, long-running calls to the model.
- Early independent tests placed K3 near the top on coding and long‑horizon tasks, yet some benchmark snapshots flagged much slower runtimes and higher per‑task costs compared with leading proprietary models.
- The launch has pushed Moonshot to speed a late‑stage fundraising push with Hong Kong listing plans and has pressured chip and AI‑linked stocks as analysts weigh how cheaper open‑weight models could change the market for cloud and inference infrastructure.