Kimi's K3 Model Features 2.8 Trillion Parameters, Releases July 27
- Published
- Jul 16, 2026 — 19:49 UTC
Kimi’s upcoming K3 model boasts 2.8 trillion parameters and a context capacity of one million tokens. Scheduled for full weight release on July 27, K3 is significantly pricier than its predecessor. In internal benchmarks, K3 approaches the performance of Claude Fable 5 and GPT 5.6 Sol, outperforming Opus 4.8 and GLM 5.2 in several tests, according to Kimi. This release follows a trend of increasing costs in AI model development, indicating a shift away from super cheap Chinese AI solutions. For practitioners, the enhanced capabilities of K3 may provide new opportunities for advanced multimodal applications. More details are available from The Decoder.
By Callan Zhang · Jul 16, 2026 · Editorial standards →
Summarised from the primary source with AI assistance under human editorial oversight. Turing Wire is not a primary source — read the original for the authoritative account.
Source: The Decoder