Section
Developers
API prices, coding plans, keys, and how to call Chinese models from outside China.
Chinese AI API prices, live, versus the latest Opus and SolOpenRouter prices for Chinese models, which flash tiers are actually cheap, and where the list price is not the lab's rate.Use Chinese models in Claude Code: the base URLs we could verifyAnthropic-compatible base URLs for Z.AI and DeepSeek, plus what we would not paste in for Kimi or Qwen.Coding plans compared: GLM's credits, and the plans we would not priceGLM publishes credits and a standard price on its own docs. DeepSeek's API page is pay-as-you-go. Other plan prices were not re-verified.What a coding session costs: flash models, flagships, and the cache assumptionAn illustrative 3 million token session, priced from the snapshot. Cache-write fees are excluded and the mix is not your repo.DeepSeek API key: top up, then use the off-peak hours on purposeThe docs bill a topped-up balance. Off-peak is half. We did not confirm which cards work from abroad.DeepSeek V4.1 Flash on OpenRouter: the cheap default, with the direct price beside itOpenRouter's V4.1 Flash card matches DeepSeek's peak rate. Off-peak direct is half. Here is the setup and the catch.GLM Coding Plan: Lite, Pro, and Max, in credits not slogansZ.AI's coding plan starts at $18 on the migration notice, with 5-hour and weekly credit caps. What the docs actually say.Is Kimi K3 open source? The weights are public. The licence is not Apache.Moonshot released Kimi K3 weights under the Kimi K3 License, with revenue thresholds for API businesses and large products.Kimi K3: official price, OpenRouter's cheaper host, and how to call itMoonshot lists Kimi K3 at $3 input and $15 output. OpenRouter's headline price is a different host. Weights are public, with a licence catch.MiMo, StepFun, and Hunyuan: cheap on the list, thin on evidenceMiMo V2.6 Pro and Hunyuan HY3 are inexpensive on OpenRouter. We do not have lab invoices or SWE-bench rows for them.MiniMax M3: the OpenRouter price, and what it is notM3 is cheap on the OpenRouter blend. That chart is not a Hailuo video price and not a SWE-bench win we could find for M3.OpenRouter versus direct: where the Chinese model is actually cheaperDeepSeek off-peak undercuts OpenRouter. GLM-5.3 and Kimi K3 list prices are cheap hosts, not the labs.Paying for Chinese AI from abroad: what the docs show, and what they do notUSD prices exist for Z.AI, DeepSeek, and Kimi. Which foreign cards clear was not tested. No reseller links.Qwen API: international sign-up, and the free tier we would not inventModel Studio and the China console are different accounts. We did not confirm a free-token number, so this page does not print one.Run Qwen locally when the API price is not the constraintSmaller Qwen ids have a public API price you can compare with a GPU you already own. We do not invent a tokens-per-second figure.Z.AI API: Flash, FlashX, and the flagship price people misreadGLM-5.3 Flash matches the docs at $0.15 / $0.50. The flagship is $1.40 / $4.40, not the OpenRouter headline.