Author(s): Caspar Bannink

Originally published on Towards AI.

China Just Shipped Opus 4.8-Level Agentic Coding for One-Sixth the Price

China just struck again on the AI release curve. A new open-weights coding model from Moonshot AI (known for K2.6 and the K2 Thinking line) ships with Opus 4.8-level agentic coding at one-sixth the price of GPT-5.5, and you can download the weights today. The model is called Kimi K2.7 Code, and on the only publicly verified coding benchmark on its launch table, MLS-Bench Lite, it scores 35.1. GPT-5.5 scores 35.5. They are within 0.4 points of each other. Claude Opus 4.8 leads at 42.8.

After introducing Kimi K2.7 Code, the article argues that its real disruption is pricing and task economics: it’s far cheaper than Opus 4.8 and GPT-5.5 on output tokens, with the weight download described in detail (595GB safetensors, 1T/32B MoE, 256K context, native INT4, Modified-MIT license). It emphasizes that while capability varies by benchmark, the model is competitively calibrated—especially on agentic coding and terminal-use—using Artificial Analysis’ Intelligence Index and eval panels as the key calibration. The piece also frames this as part of a broader “China-AI” catch-up trend (citing Nex-AGI’s Nex-N2-Pro as the prior week’s data point), then walks readers through how to validate the claim themselves: run K2.7 on vLLM with representative tasks, measure success and cost per accepted change, compare against a frontier closed model under the same agent loop, and—if needed—rent data-center hardware because “open weights” still implies serious compute.

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