Local AI benchmarks
Every model gets the same prompts on the same hardware. Pick any two to compare their speeds and the pages they wrote, running live in your browser.
Get Ownlight
Every number here comes from Ownlight, the workspace that runs Claude Code, Codex and local models side by side.
Compare two models
Run your own agents on models like these.
Ownlight puts Claude Code, Codex, opencode and DeepSeek Harness side by side on one canvas, and can point them at an Ownlight server or Ollama on your own hardware. Local sessions show their generation speed live.