Skip to main content
fireconnect claude demo sends the same code-generation prompt to two models through your current FireConnect Claude profile. It shows both runs together so you can compare speed, estimated cost, and output. Each model runs in a separate temporary working directory. The demo does not modify ~/.claude/settings.json.

​
Compare the results

  • Live terminal race: Watch both models stream code with token counts and wall-clock timers.
  • Browser comparison: Run both generated apps side by side and compare measured speed and estimated cost.
  • Saved output: Find compare.html, result.json, generated apps, and logs in ./fireconnect-demo/.

​
Compare the generated apps

FireConnect demo browser comparison page with two Tetris apps side by side and speed and cost metrics

Browser comparison from a real FireConnect demo run: two Tetris apps with measured speed and estimated cost

​
Watch both models run

Animated capture of a FireConnect split-pane terminal race between Claude Sonnet and GLM 5.2 Fast on a Tetris prompt

Real FireConnect demo race: GLM 5.2 Fast finishes while Claude Sonnet is still running

​
Requirements

  • FireConnect installed
  • Claude Code available as claude on your PATH
  • Claude Code connected with fireconnect claude
The demo supports Claude Code only. Both sides use the authentication and routing from your active FireConnect profile.

​
Run the demo

That runs the default Tetris preset: both sides get the same prompt to build a playable game in a single HTML file.

​
Choose a prompt

Built-in presets are tetris, snake, and clock. Tetris is the default. Use --prompt for custom text or --prompt-file to read a task from a file.

​
Choose models

The defaults are opus on the left and glm-fast-latest on the right. Here, opus is a Claude Code alias, not a Fireworks model ID. You can use any model or Claude alias available in the connected profile.

​
Automate a race

​
Clean up

demo clean deletes an output directory only when it contains a demo marker such as result.json or compare.html. It leaves unrelated directories untouched.

​
How results are produced

  • Each side runs claude -p in a separate temporary directory with the same prompt and active FireConnect profile. Only the model changes.
  • Speed comes from the measured run. Estimated cost uses token counts and the published rate for each model.
  • If a model fails, the comparison reports the failure instead of declaring a winner.
  • compare.html contains both apps and works offline.