Both are in the catalog already
If you have a ClawCloud instance on managed credits, DeepSeek and Qwen models are available without any extra configuration. They're part of the 100+ curated models routed through OpenRouter.
Send /model followed by the model ID to switch:
/model deepseek/deepseek-r1
/model qwen/qwen3-max
The bot confirms the change and uses it for all following messages.
DeepSeek models on ClawCloud
Three DeepSeek models are in the managed catalog:
| Model ID | What it's for |
|---|---|
deepseek/deepseek-r1 | Reasoning. Works through problems step-by-step before answering. Good for math, logic, and code debugging. |
deepseek/deepseek-v3.2 | General chat. Fast, capable, cheaper per token than most Claude or GPT options. |
deepseek/deepseek-chat-v3.1 | Previous-generation chat model. Still solid for straightforward conversations. |
There's also a free-tier option: deepseek/deepseek-r1-0528:free. It costs zero credits and is a good way to try DeepSeek's reasoning model without using your budget.
Qwen models on ClawCloud
Four Qwen models are available:
| Model ID | What it's for |
|---|---|
qwen/qwen3-max | General-purpose with a large context window. Handles long conversations and documents well. |
qwen/qwen3-235b-a22b | The largest Qwen model. Uses mixture-of-experts architecture. |
qwen/qwen3-coder | Code-focused. Better at programming tasks than the general models. |
qwen/qwen3-coder-plus | Enhanced version of Qwen3 Coder with higher quality on complex code. |
Free-tier option: qwen/qwen3-coder:free. Zero credits for a code-focused model.
Cost comparison
DeepSeek and Qwen are generally cheaper per token than Claude Sonnet or GPT-4.1. If you're on a managed credits plan and watching your usage, switching to DeepSeek V3.2 for routine conversations and reserving a premium model for harder tasks can stretch your credits further.
The free-tier variants (deepseek/deepseek-r1-0528:free and qwen/qwen3-coder:free) cost nothing against your credit balance. They run at lower priority and may have rate limits, but they work for testing or light use. See the credits guide for details on how credit consumption works.
BYOK options
If you bring your own key instead of using managed credits, you can access DeepSeek and Qwen through several paths:
- OpenRouter key: All the models listed above are available since ClawCloud's managed catalog uses OpenRouter.
- Qwen native: OpenClaw has a built-in Qwen OAuth plugin (
qwen-portal-auth) that provides free-tier access to Qwen Coder and Vision models directly from Alibaba Cloud. - DeepSeek via Hugging Face:
huggingface/deepseek-ai/DeepSeek-R1is available through Hugging Face Inference if you have a HF API key.
See the OpenRouter model backend guide and the OpenClaw model providers docs for setup details.
When to use which
- Reasoning tasks (math, code review, debugging): DeepSeek R1
- Daily chat on a budget: DeepSeek V3.2
- Code generation: Qwen3 Coder or Coder Plus
- Long documents or context-heavy work: Qwen3 Max
None of these replace Claude or GPT for every task. Claude Sonnet 4 and GPT-4.1 are still strong general-purpose choices. But if you want to try something different, or your credits are running low, DeepSeek and Qwen are solid alternatives already on your OpenClaw server.
For the full model list and aliases, see OpenClaw Models and the model switching guide.