Averaged across all 25 tasks on this site, for all 7 models. Same method as every other page here: recorded runs where they exist, clearly-labelled placeholder profiles where they don't — never a marketing number.
Averaged across every task and every model on each side: OpenAI's models come out faster overall (516× vs 717× human speed, Anthropic vs OpenAI). Accuracy averages close, 84% vs 79%, and cost averages $0.01 vs $0.0061 per task. But that average hides real variation — the fastest single model here is GPT-5.6 Luna (1,240× on average), and tier-for-tier the gap narrows or flips.
Averages are a starting point, not a verdict: Compare AI Models lets you race any specific Claude model against any specific GPT model on one task, at one effort level, and see the real numbers for that exact pairing.
Ranked by average speed multiplier over a typical person, across all 25 tasks:
Ranked by average cost per task (same tasks, same token-count method, different price per token per model):
The most expensive model here averages roughly 28× the cost of the cheapest, for the same set of tasks — that gap is the entire reason this site shows tokens and cost on every result, not just speed.
Ranked by average accuracy against each task's scoring rubric:
The fastest model and the most accurate model are rarely the same one — see the FAQ for why speed, cost, and accuracy are three separate facts about a model, not one score.
This site's own recommendation logic, per tier and provider — the same "good fit" judgment shown on every result:
All figures are averages of this site's own per-task numbers — recorded weekly where a run exists, a clearly-labelled placeholder profile otherwise. See how we measure →. Want one specific pairing instead of an average? Discover AI Speed races one task at a time; Compare AI Models runs two models head-to-head.