How do I use the DeepSeek, Qwen, Kimi and GLM API with one key at a flat monthly price?

Last updated October 26, 2026 by the APIClaw team

APIClaw gives you one OpenAI-compatible key for DeepSeek, Qwen, Kimi and GLM models at a flat monthly price from $19 to $129. You set the base URL to https://apiclaw.biz/v1, pick a model ID such as deepseek-v4-pro or kimi-k3, and each request counts against a daily quota instead of a per-token bill.

One key for four Chinese model families

DeepSeek, Qwen, Kimi and GLM are the open-weight and China-based model families that many developers now use for everyday coding work. They are fast, they handle long context well, and they are often much cheaper per token than the biggest Western models. The friction is on the account side: four vendors usually means four sign-ups, four API keys, four billing dashboards and four sets of rate limits.

APIClaw puts all of them behind one OpenAI-compatible endpoint. You create one key, point your client at https://apiclaw.biz/v1, and choose the model by its ID. The same key also reaches Claude, GPT, Gemini, Mistral, Llama, Cohere and Grok, so you can switch models inside a session without touching your credentials. APIClaw routes requests to models available through publicly available sources and provider APIs, and it is independent: it is not affiliated with Anthropic, OpenAI, Google or Cursor.

The model IDs and what each request costs

Every plan counts requests, not tokens. Each model has a weight, which is the number of requests one call uses from your daily quota. Here are the Chinese-model IDs you can call today.

The model hub always shows the current list and weights, so check it before you hard-code an ID. Both DeepSeek models count as a single request, which makes them the cheapest way to spend a daily quota. The other families count as two.

What the quota looks like in practice

Plans start at $19 a month for 500 requests per day and go up to $129 a month for 8,000 requests per day. The quota resets every day at 00:00 UTC. There is no 5-hour window, no weekly cap and no per-token charge, and there is no limit on tokens per request beyond the model's own context window, which reaches 1M tokens on many models.

Plans run in 30-day cycles and you can cancel at any time. If you need more than 8,000 requests a day, ask for a custom plan through Telegram support. You can mix models freely: the weights simply add up against the same daily number.

Set it up in Qwen Code

Qwen Code reads an OpenAI-compatible endpoint from three environment variables. Export them in your shell, then start the tool:

export OPENAI_API_KEY="YOUR_APICLAW_KEY"
export OPENAI_BASE_URL="https://apiclaw.biz/v1"
export OPENAI_MODEL="qwen3.8-max"
qwen

To switch to a cheaper request weight for routine edits, change OPENAI_MODEL to deepseek-v4-flash. The full walkthrough, including how to keep the settings in a project file, is in the Qwen Code setup guide.

Set it up in OpenCode

OpenCode takes custom providers in its opencode.json file. Add an OpenAI-compatible provider that points at APIClaw and list the models you want:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "apiclaw": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "APIClaw",
      "options": {
        "baseURL": "https://apiclaw.biz/v1",
        "apiKey": "YOUR_APICLAW_KEY"
      },
      "models": {
        "deepseek-v4-pro": { "name": "DeepSeek V4 Pro" },
        "kimi-k3": { "name": "Kimi K3" },
        "glm-5.3": { "name": "GLM 5.3" },
        "qwen3.8-max": { "name": "Qwen3.8 Max" }
      }
    }
  }
}

After saving, pick the model from the OpenCode model list. More detail, including how to store the key outside the config file, is in the OpenCode setup guide.

Test it with curl first

Before wiring up an editor, confirm that your key works with one request:

curl https://apiclaw.biz/v1/chat/completions \
  -H "Authorization: Bearer YOUR_APICLAW_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [{"role": "user", "content": "Write a Python function that reverses a linked list."}]
  }'

The response follows the standard Chat Completions shape, so any SDK that speaks the OpenAI protocol works unchanged. Swap the model value for kimi-k3, glm-5.3 or qwen3.8-flash to compare answers on the same prompt.

Choosing between the four families

There is no single winner, and the honest answer is to test on your own code. A practical way to start: use a DeepSeek model for high-volume work such as refactors, test generation and code explanation, because it costs one request per call. Move to Qwen or Kimi when you want a second opinion on a hard problem or a long-context task. Use GLM when you want a lighter, quick model with the flash variant. Because switching is only a model ID, you can keep the same agent configuration and change the ID per task.

If you also want Claude or GPT for the hardest tasks, the same key covers them. Claude Sonnet counts as 2 requests, and Claude Opus and GPT-6 Sol count as 3. The page on one API key for all models explains how the single key works across providers.

Using these models in other tools

The same base URL and key work in any tool that accepts an OpenAI-compatible provider. Cline, Kilo Code, Continue and Roo Code all have a custom provider setting where you paste the base URL https://apiclaw.biz/v1, your key and a model ID. Cursor agent mode uses https://apiclaw.biz/cursor, and in Cursor the model IDs take the apiclaw/ prefix, for example apiclaw/deepseek-v4-pro. Claude Code speaks the Anthropic Messages format, so it uses https://apiclaw.biz with no /v1 at the end. The Cursor guide and the Claude Code guide list the exact fields.

A useful habit is to keep one cheap default and one strong fallback. Set the default to a weight-1 DeepSeek model so routine edits barely touch your quota, and switch to a weight-2 model only when the first answer is not good enough. Over a month this keeps the daily quota for the tasks that need it.

When this is not the best fit

A flat plan makes sense when your usage is steady and heavy. At low volume, a few dozen modest calls a day, per-token billing is often cheaper than $19 a month, and going direct to each vendor or through a pay-as-you-go router may cost less. If your work is mostly long agent sessions, the flat price tends to win because the tokens per request do not change the bill. For a side-by-side view of how APIClaw, OpenRouter, direct APIs and Claude plans compare, see the comparison page, or read the OpenRouter alternative page if you currently pay per token.

Get started in a few minutes

Sign up for the free trial: 50 requests, no card needed, usable across the models in the table above. Payment for paid plans is in crypto through NOWPayments, covered on the pay with crypto page. The docs home has setup guides for Claude Code, Codex CLI, Cursor and more, so you can use the same key in every tool you already run.