How fast Claude and OpenAI models answer the same real-world task, measured every hour from TraxAgent. Median of the last 24 hours.
Each model gets the identical prompt once an hour, streamed, from a single region. Time to first token is from sending the request to the first visible word. Output speed counts output tokens after the first token, including reasoning tokens where the provider bills them as output. Cost per task uses list prices. Your own traffic may differ with prompt length, region and settings.
Want this for your own prompts and models? TraxAgent's model comparison runs your real prompts on 2 or 3 models and scores quality, speed and cost per task. See plans.