Moonshot introduced Kimi K3, a multimodal model with 2.8 trillion total parameters. Its launch materials emphasize long coding sessions, visual reasoning, frontend development, and analysis of large collections of information.
Kimi describes frontier-level results across its evaluation suite while acknowledging that overall performance still trails the strongest proprietary models, Claude Fable 5 and GPT-5.6 Sol. A lead on a particular frontend or coding benchmark does not mean it leads every kind of task.
Standard API prices at launch are $3 per million uncached input tokens and $15 per million output tokens; cached input is $0.30 per million tokens. The company planned a public weight release after launch.
Its large total parameter count is a milestone for open models, but deploying the full model is a substantial infrastructure task. See the official Kimi K3 blog for architecture, availability, and evaluation details.
Adapted from the original Chinese article, published on July 18, 2026.
Comments NOTHING