Kimi’s open model K3 nears GPT-5.6 Sol and Fable 5 while signaling the end of super cheap Chinese AI
What happened
Kimi announced K3, a new multimodal AI model with 2.8 trillion parameters and a context window of one million tokens. Early benchmarking by Kimi shows K3 performing close to OpenAI’s GPT 5.6 Sol and Anthropic’s Claude Fable 5 models, surpassing competitors like Opus 4.8 and GLM 5.2 by a significant margin. K3’s full weights are set to be released by July 27. The model also comes with a notably higher price than Kimi’s earlier offerings.
Why it matters
K3’s scale and performance push the boundaries of accessible open-weight models, narrowing the gap with industry-leading commercial AI. The massive context window and multimodal capabilities suggest improved utility for complex tasks requiring long-term memory or cross-modal understanding. However, the higher cost signals a shift away from the era of ultra-cheap Chinese AI models flooding the market. This raises the operational cost floor for builders relying on affordable open models, potentially impacting startups and researchers who depend on low-priced options.
What to watch next
How other open model creators respond to K3’s leap in size, cost, and performance will influence the competitive landscape. Whether Kimi’s bigger model will become standard for demanding applications and how pricing trends evolve will determine affordability for smaller teams. Also, the quality of K3 in real-world settings versus benchmarks and the openness of its release will affect adoption and downstream innovation. The upcoming full release on July 27 will be a key moment for builders and investors tracking open AI models.
AI Quick Briefs Editorial Desk
