LM-Speed-X

LM-Speed-X is an LLM API speed testing tool, forked from nexmoe/lm-speed, supporting speed tests for OpenAI-compatible APIs, DeepSeek, Silicon Flow, Tencent Cloud, and more.

Try it Online

Visit lm-speed-x to use it online

The original tool is feature-rich but requires a PostgreSQL database to run, making deployment inconvenient. I made a series of improvements to make it more accessible:

• Database-free: Switched data storage to browser localStorage, no database deployment required, significantly lowering the barrier to entry

• Local Rankings: Added local speed test ranking with custom filters, data export/import for easy cross-API performance comparison

• Config Import/Export: One-click export/import of test configurations for easy sharing across devices

• URL Param Quick Test: Start a speed test instantly via URL parameters (baseUrl, apiKey, modelId) — no manual form filling needed

  Format:

  https://lm-speed-x.xtsat.cc.cd/?baseUrl=<YOUR_BASE_URL>&apiKey=<YOUR_API_KEY>&modelId=<YOUR_MODEL_ID>

  Example with DeepSeek official API:

  https://lm-speed-x.xtsat.cc.cd/?baseUrl=https://api.deepseek.com/v1&apiKey=sk-xxx&modelId=deepseek-v4-pro

  All three parameters are required. The page will start the speed test automatically.

The test uses a five-round continuous stress testing mechanism with tiktoken for precise token counting, covering key metrics such as first-token latency and response time, and automatically generates performance analysis reports.

For local deployment, simply fork the repo and run pnpm dev to start a local server accessible in your browser.