The best new coding model this week is Chinese, but it isn't cheap
Moonshot AI released Kimi K3 on 16 July: a 2.8-trillion-parameter open-weight model, API live now through OpenRouter and Moonshot's own endpoint, with full weights due 27th July. It debuted at #1 on LMArena's Frontend Code Arena and beats GPT-5.6 and Fable 5 on some benchmarks. The twist is the price. At $3 per million input tokens and $15 output, K3 sits at exact parity with Claude Sonnet, the most expensive model any Chinese lab has ever shipped, and five times the cost of Moonshot's own K2 family.
What it actually means
For a month the only reason to reach for a Chinese model here was price , swap the config line, cut the bill. Kimi K3 kills that framing. It isn't undercutting anyone; it's charging Sonnet money and topping the frontend coding board to justify it. The shift matters more than the model. While these were the budget option, the decision was easy and low-stakes: try it on throwaway work, keep the expensive model for the parts that matter. K3 removes the easy out now the open Chinese model might be the best tool for the part that matters, at full frontier price. The budget alternative just became a real contender.
The Chinese model to beat is no longer the cheap one — it's the best one.
$3/$15Kimi K3's pricing, level with Claude Sonnet and a first for a Chinese lab
Source: kimi.com · Friday 17 July 2026