Skip to content

Speed

Speed measurements are coming soon. We will measure speed ourselves by streaming a fixed prompt and timing the response, and publish medians across trials.

No measurements published yet

Speed is only shown when we have measured it ourselves; we do not copy speed figures from other sites. To publish measurements, run the speed runner with an OpenRouter key and rebuild:

python pipeline/speed.py --top 20 --trials 3 --max-usd 1.00 --dry-run   # see the plan and worst-case cost
OPENROUTER_API_KEY=... python pipeline/speed.py --top 20 --trials 3 --max-usd 1.00
python pipeline/build.py

Each trial records time to first token (any token, including reasoning), time to the first answer token, total time and output tokens. Output speed = output tokens ÷ (total time − time to first token).