SemiAnalysis: Kimi K3 ranks third globally in comprehensive strength, overseas leading AI may hide high profit margins

Odaily Planet Daily reported that semiconductor research firm SemiAnalysis analysts Jordan Nanos and Max Kan released a report interpreting Moonshot AI's Kimi K3. Comprehensive benchmark tests show that Kimi K3, with 2.8 trillion parameters, ranks third globally in overall capability, trailing only Fable 5 and GPT-5.6, while outperforming Google Gemini, directly illustrating the continuously narrowing AI gap between China and the US.

Analysts pointed out that Kimi K3 cannot be deployed on a single Nvidia B200 GPU, requiring high-performance computing hardware such as GB300 or AMD MI355X. This suggests that the flagship model parameter sizes of OpenAI and Anthropic are similar, with no generational magnitude lead.

In terms of pricing, Kimi K3 charges $3 per million tokens for input and $15 per million tokens for output, aligning with Anthropic Sonnet, and represents a threefold price increase over its predecessor. Analysts extrapolate that if Moonshot AI can operate healthily, the higher pricing of OpenAI and Anthropic implies that their API business profit margins are significantly higher than those of SaaS. This conclusion is solely based on reverse calculations from Kimi's parameters, performance, and pricing, and is not derived from publicly available financial reports.