Kimi K3 Ranks Third Globally, SemiAnalysis Reveals Potential Profit Margins for OpenAI and Anthropic

According to SemiAnalysis analysts Jordan Nanos and Max Kan on July 20, Kimi K3—developed by Moonshot AI—ranks third globally in comprehensive benchmarks, surpassing Google Gemini and trailing only Fable 5 and GPT-5.6. The model features 2.8 trillion parameters, requiring high-end hardware such as NVIDIA GB300 or AMD MI355X systems for deployment.

Kimi K3's pricing aligns with Anthropic's Sonnet series at approximately $3 per million input tokens and $15 per million output tokens, representing a threefold increase from its predecessor. Based on these figures and the model's performance, the analysts inferred that competing closed-source models from Anthropic and OpenAI likely operate at similar parameter scales. Max Kan noted that if Moonshot operates profitably at current pricing, rivals charging higher rates for comparable models suggest their API businesses may sustain substantial profit margins.