Moonshot AI’s Kimi K2.6 has become the top-ranked open-weights model, landing fourth globally and closing to within three points of the leading US frontier models, signalling a major shift in open-source AI competitiveness.
Moonshot AI has pushed open-source AI closer to the frontier with Kimi K2.6 debuting at fourth on the Artificial Analysis Intelligence Index v4.0, scoring 54 points—just three behind Anthropic’s Claude Opus 4.7, Google’s Gemini 3.1 Pro and OpenAI’s GPT-5.4, all tied at 57.
The ranking makes Kimi K2.6 the highest-performing open-weights model and underscores how rapidly Chinese open-source models are closing the gap with proprietary US systems. It also leads the next-best open model, GLM-5.1, by three points.
The index aggregates 10 evaluations spanning reasoning, coding, general knowledge and agentic performance, where K2.6 made its sharpest gains. Its GDPval-AA Elo jumped to 1520 from 1309 in K2.5, while a 96% τ²-Bench Telecom score placed it firmly among frontier models.
Enterprise credibility was strengthened by a reduced hallucination rate of 39%, down from 65% in the prior generation, approaching proprietary frontier benchmarks.
Built on a Mixture-of-Experts architecture with one trillion parameters, 32 billion active parameters and a 256k token context window, K2.6 also consumed about 160 million reasoning tokens in benchmark runs, reflecting substantial compute intensity.
With broad third-party availability and sustained adoption signals, the result strengthens the view that Chinese open-source AI is becoming a serious challenger to closed frontier labs.
