AI Model Benchmarks

368 scores across 10 benchmarks covering quality, coding, reasoning and speed, each with its source.

Showing current models: released in the last 12 months, and not replaced by a newer version.

Output tokens per second over a whole response, counting the wait for the first token. Averaged across the test workloads Artificial Analysis ran in the past 5 days. Higher = faster responses. Source: Artificial Analysis, which measures every model the same way.

Model
tok/s

Every score links to its source and shows when it was measured or published. Independent evaluators (Artificial Analysis, Vals AI, LMArena, Vectara) run the tests themselves; scores marked as from a developer are that company's own claim, and evaluators often measure the same model lower. Scores with no traceable source are not shown. Higher is better for all metrics except Hallucination Rate, where lower is better.