Compare model releases
Compare two model releases at the same reasoning effort, with sources for every reading.
Select the model you use and the release you want to compare. Missing effort levels remain visible.
Sources and limits
Test columns marked AA are Artificial Analysis runs, not official agent leaderboard submissions. Measurement dates are not supplied by the connected API.
The connected API does not supply current task cost, time or output tokens. History includes older source captures with their original clock and unknown timezone. They do not replace current readings or token list prices.
Official agent runs retain the agent, benchmark version, model, effort and run identity. They do not contribute to AA indexes.