AI Models
1629Latest models
At a glance
Claude Fable 5.2 (Default Fallback) leads with 64.6 (Intelligence Index), out of 581 ranked models. Best score for the money: Agnes 3.0 Flash, 35.5 for $0.075 per million tokens. Fastest: Mercury 2.5, 532 t/s.
Value and speed are picked among the top 291, so a cheap or quick run with a low score never wins. Data retrieved on Nov 22, 2026.
0 results of 581
Search also includes models without scores, shown without a rank. The full catalogue is available in Advanced mode.
No models match your search.
What is BenchSift?
BenchSift is a free, independent catalogue for choosing an AI model. It brings together measurements that are usually spread across several sites: Artificial Analysis benchmarks (intelligence, coding, math, speed and latency), OpenRouter prices, capabilities and usage rankings, Models.dev specifications and the open-weights evidence published on Hugging Face. Every figure links back to its source, and a value nobody has published stays empty instead of being estimated.
It is built for developers, product teams and curious readers who need a model for a real job: a coding agent, a chat assistant, a document pipeline or a high-volume task on a budget. Start from the ranking above, sort and filter it, open a model page to see where it ranks on each benchmark, browse every model or put up to four models side by side in the comparator.
Where many comparators show a single leaderboard or vendors' own claims, BenchSift keeps each source's methodology visible, groups the reasoning levels of one model on a single row and gives their own pages to benchmarks the usual tables leave out, such as DeepSWE, FrontierCode or the Cyber Index. The whole site is available in English and French; the About page details the sources.