Each provider's strongest model, shown at Medium effort by default (switch to High, Max or each model's best score above; the header numbers, chart and analysis follow this setting). Intelligence is the Artificial Analysis Intelligence Index. Speed is output tokens per second once the model starts answering. Time per task is the whole wait for a 500-token answer, thinking included, measured by Artificial Analysis on real ~10k-token requests (median of the last 72 hours). It depends hugely on the effort setting: at Max the top Claude and GPT models think with almost no budget limit, so a request that takes ~10–20 seconds at Medium or High can take several minutes at Max. Use the Effort switch, or hover a row to see every setting.