DeepSeek and Qwen in Cursor — from China to production
Developers in China can connect Cursor, Claude Code, and Windsurf to DeepSeek R1 and Qwen3 models via SiliconFlow through TheRouter. SiliconFlow runs on Aliyun for low latency from mainland China. DeepSeek R1 reasoning_content is preserved end-to-end. Cost is 40–85% lower than US-based providers. One API key covers all models.
Chinese developers can connect Cursor, Claude Code, and Windsurf to DeepSeek R1, Qwen3 Coder, and 50+ other models through TheRouter. SiliconFlow's Aliyun routing gives you low-latency access to the best open-weight models at a fraction of Bedrock's cost.
Best models for vibe coding via SiliconFlow
DeepSeek R1
deepseek/deepseek-r1Best for complex reasoning, debugging, architecture decisions. Exposes chain-of-thought in reasoning_content.
Qwen3 Coder 480B
qwen/qwen3-coder-480bAlibaba's largest coding model. Specialized for code generation, refactoring, and multi-file edits.
DeepSeek V3.2
deepseek/deepseek-v3.2Fast, cost-efficient general-purpose model. Great for code completion, documentation, and chat.
Qwen3 Coder 30B
qwen/qwen3-coder-30bLightweight coding model for fast completions. Good for autocomplete and small edits.
Connect Cursor in 2 minutes / 两分钟接入 Cursor
1. Get your API key
Sign up at app.therouter.ai and copy your API key.
2. Configure Cursor
Open Cursor → Settings → Models → Add Custom Model:
Base URL: https://api.therouter.ai/v1 API Key: <your TheRouter key> Model: deepseek/deepseek-r1
3. Start coding
Select deepseek/deepseek-r1 (or any other model) in Cursor. TheRouter routes through SiliconFlow for China-optimized latency and 84% cost savings.
Common questions
Which coding tools work with TheRouter for DeepSeek and Qwen?
Cursor, Claude Code, Windsurf, Cline, Continue, and any tool that supports an OpenAI-compatible base URL. Set the base URL to api.therouter.ai/v1 and your TheRouter API key.
Does DeepSeek R1 work in Cursor with chain-of-thought visible?
Yes. TheRouter preserves the reasoning_content field from DeepSeek R1 end-to-end when routing through SiliconFlow. Coding tools that display thinking steps will show the full chain-of-thought.
Why use SiliconFlow for coding in China?
SiliconFlow runs DeepSeek and Qwen models on Aliyun infrastructure in China, so latency is significantly lower for developers in mainland China than routing through US-based Bedrock. Cost is also 40–85% lower.
Can I switch between DeepSeek and Claude without reconfiguring Cursor?
Yes. Configure Cursor with TheRouter once. Then switch models by changing the model name in Cursor settings — claude-sonnet-4-6, deepseek/deepseek-r1, qwen/qwen3-235b, etc. TheRouter handles the routing transparently.