Author(s): Kashif Mehmood

Originally published on Towards AI.

China’s Coding Model, Kimi K2.7 Code, is 6x Cheaper Than Claude. It also Grades Its Own Homework

China’s best open-weight coder is here; it’s a fraction of the cost, and you cannot independently verify a single number it ships with. That gap, not the capability one, is the whole story now.

The article argues that Kimi K2.7 Code’s standout value is not just its low price but the way its results can’t be independently checked: Moonshot publishes mostly proprietary benchmarks it controls, with no entries on major public leaderboards like SWE-bench Verified/Pro, Terminal-Bench, or LiveCodeBench. It contrasts vendor-reported internal scores—where K2.7 “wins” against its predecessor and often loses less than expected—with the closed frontier’s publicly rerunnable numbers (e.g., Opus 4.8’s and GPT-5.5’s well-sourced performance). The author details why the model can be cheaper (mixture-of-experts routing plus fewer “thinking” tokens), what “open” really means in practice (size, hardware needs, and an Anthropic-compatible API endpoint), and what to scrutinize before trusting claims (missing public benchmarks, unverifiable proprietary evals, and reports of imperfect real-world behavior). Overall, it suggests China’s capability gap is shrinking, but the trust/verification gap is still decisive for anyone choosing models for production use.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI