Kimi K2.6, a Chinese-made AI model with performance equivalent to Claude Opus 4.6, has been released as an open model.



Chinese company

Moonshot AI announced its AI model ' Kimi K2.6 ' on April 21, 2026 (Japan time). Kimi K2.6 has achieved scores exceeding Claude Opus 4.6 in multiple benchmark tests and is available as an open model that anyone can download.

Kimi K2.6 Tech Blog: Advancing Open-Source Coding
https://www.kimi.com/blog/kimi-k2-6




Kimi K2.6 is a MoE model with a total of 1 trillion parameters and 32 billion active parameters. It boasts superior coding and agent performance.

The following diagram shows the benchmark results for 'Kimi K2.6', 'GPT-5.4 (xhigh)', 'Claude Opus 4.6 (max effort)', and 'Gemini 3.1 Pro (thinking high)'. Kimi K2.6 achieved the highest scores in Humanity's Last Exam, which measures agent performance, and SWE-Bech Pro, which measures coding performance.



Kimi K2.6 can also perform long-term tasks. When tasked with 'optimizing Qwen3.5-0.8B inference code written in Zig for a Mac with M3 Max,' it ran for over 12 hours, including more than 4000 tool calls, and successfully achieved optimizations that surpassed LM Studio's inference speed.



Kimi K2.6 is also characterized by its excellent aesthetic sense, and can automatically generate visually appealing websites.



In an internal comparison of the design performance of the Kimi K2.6 and Gemini 3.1 Pro at Moonshot AI, the Kimi K2.6 won 47.5% of the time.



Kimi K2.6 is available at the following link.

moonshotai/Kimi-K2.6 · Hugging Face
https://huggingface.co/moonshotai/Kimi-K2.6



It can also be used via API , with a price of $0.95 (approximately 151 yen) per 1 million tokens for input and $4 (approximately 636 yen) for output.

Related Posts:

in AI, Posted by log1o_hf