Kimi K3: China's New AI Challenges OpenAI & Anthropic

China now has a new strong AI model called Kimi K3, developed by Moonshot AI. This model aims to compete with leading AI from Anthropic and OpenAI. Moonshot AI describes Kimi K3 as one of its most capable models yet. While it hasn't surpassed the very best proprietary AI models, the company believes the gap is closing. Here's the thing: Kimi K3 performed competitively with Fable 5 in GPU kernel optimization and substantially outperformed Anthropic's Opus 4.8, GPT 5.6 Sol, and GPT 5.5 in this area. On general AI agent benchmarks, Claude Fable 5 leads in overall reasoning with a score of 1,760 on GDPval-AA v2 Elo, ahead of GPT-5.6 Sol at 1,748 and Kimi K3 at 1,668. However, Kimi K3 scored 77.8 on the Program Bench, narrowly beating GPT-5.6 Sol at 77.6 and Claude Fable 5 at 76.8. It also topped SWE Marathon with a score of 42.0. What's interesting is that Kimi K3 is the first open-weight AI model to approach the three trillion-parameter mark. It also features a one million-token context window. The bottom line is that the Kimi K3 shows the global AI race for leadership is no longer just among American companies.