Gate News message, April 27 — Moonshot AI plans to launch Kimi K3 in Q3 2026 (July-September), according to reports. The model will feature over 2.5 trillion parameters, with internal testing revealing context window capabilities exceeding 1 million tokens.
However, whether the 1M context will be available in the public release remains uncertain. The primary bottleneck limiting the 1M context rollout is computational resources rather than technical capability, sources indicate. Kimi’s current public version K2.6 maintains a 256K context window.
The development comes as DeepSeek V4 has already positioned 1M context as a public offering, raising the bar for long-context capabilities in open-weight models. If Kimi K3 brings 1M context to its public release, Moonshot would match DeepSeek V4 in both model scale and long-context performance.
Related News
TSMC “Sprint to 1nm” vs. Samsung “Consolidate 2nm,” the two leading chip foundries diverge on strategy
Guo Ming-chi: OpenAI wants to build an AI Agent phone; MediaTek, Qualcomm, and Luxshare Precision are key in the supply chain
Toku Launches Makimoto Open-Source AI for Asia-Pacific