Kimi Releases K2.6 Code Preview: Benchmarks, Pricing and Early Reactions

Today Kimi rolled out a new model: K2.6-code-preview.

Unexpectedly, because during the closed testing phase there was almost no promo, and mentions from users started appearing right before the release. There isn’t even a detailed technical description on the website.

But the community reaction is noticeable. Some say the level is like Claude Sonnet 4.6, others say the reasoning style is similar to Opus.
And some even say it’s the #1 open model for agentic development

What the benchmarks show according to community data:

— overall score: from 83 → 89
— SWE-Bench Verified: 76.8%
— LiveCodeBench: 85%

Overall, it’s rated at the level of Sonnet 4.6. If this is confirmed, it’s a serious step forward for Chinese models in coding.

But it’s important: this is only a preview. There are no full technical details, and the data is mostly from community tests. Final conclusions — only after the release.

The price is nice: Kimi Code offers a subscription starting at 39 yuan per month with access to K2.6.

Against the backdrop of Claude’s API pricing, this looks significantly cheaper.

For programmers, it’s a great price-to-capability ratio, and everyone else can just watch for now.