Chinese artificial intelligence lab Moonshot AI on July 16 released its K3 model, which exceeds or nearly matches the capabilities of U.S. AI companies' most advanced publicly released AI models.

Artificial Analysis' Artificial Analysis Intelligence Index, which ranks global AI models based on nine different AI evaluation benchmarks, ranked K3 in third place, slightly behind Claude Fable 5 and GPT-5.6 Sol. K3 also debuted as the top-ranked model on Arena AI's code rankings, ahead of both Fable and Sol. Its cost per task and per token is also materially lower than those of both Sol and Fable. These evaluations exclude Anthropic's Mythos AI model, which has not been publicly released. Chinese company Z.ai released its GLM 5.2 model in June, and cybersecurity researchers say its cyber vulnerability-finding capabilities are similar to those of Anthropic's models. Like K3, GLM 5.2 is an open-weight model.