How fast are GPT and Claude today?
API output speed, time to first token and Intelligence Index for OpenAI and Anthropic models, recorded every day from Artificial Analysis data.
Latest
Speed vs. intelligence, latest day
How to read this
These are API numbers. Artificial Analysis measures each model on the official API with its own fixed prompts (about 1k input tokens). Speed inside subscription products such as ChatGPT, Codex or Claude Code can be quite different.
Compare tokens per second within a vendor. OpenAI's tokenizer fits more text into each token than Anthropic's, so the same tok/s means more text from a GPT model.
Each day's value is the median of that day's snapshots (UTC). Artificial Analysis refreshes its numbers roughly every 3 hours and this page checks every hour. Artificial Analysis's own figures are medians over its recent measurements, so day-to-day moves are smoothed.
Reasoning effort follows Artificial Analysis's labels. Models with a single setting appear under every effort.