Kimi K3 is Moonshot AI's 2.8-trillion-parameter flagship model, released July 16, 2026, and the first open-weight model in the three-trillion class, built for developers and enterprises creating coding agents, research tools, and long-context applications. It is closer to an agent engine than a chat assistant, designed to use tools and finish complex work with less supervision.
Core Features
- 2.8T total parameters in a sparse Mixture of Experts that activates only 16 of 896 experts per token for efficiency.
- A 1,048,576-token context window via Kimi Delta Attention, which stays efficient on very long sequences.
- Native vision reads screenshots and refines visual output in the loop rather than via a bolted-on module.
- Always-on thinking mode reasons step by step before producing a final answer.
- Open weights promised by July 27, 2026 under a Modified MIT license for self-hosting and fine-tuning.
Use Cases
- Long-horizon coding: navigating large repositories, running terminals, and frontend or game work.
- Document-heavy research: spreadsheets, slides, and interactive data stories from big source sets.
- Teams wanting an open model that rivals Claude Fable 5 and GPT-5.6 Sol on coding benchmarks.
Pricing
API access costs 3 dollars per million input tokens and 15 dollars per million output tokens, with cached input at 0.30 dollars. In the Kimi app, K3 Max starts at a 199 yuan per month tier. Consumer sign-ups were briefly paused in July 2026 after demand spiked past capacity.
Our Take
Best open-weight option for serious coding and agentic work if you can self-host or pay API rates. The catch is that it still trails Claude Fable 5 and GPT-5.6 Sol overall, and hosted serving is pricey next to smaller models. Explore more at aifreetool.site/tool-category/ai-engine-model/.
