Artificial Analysis
An independent site that benchmarks AI models and the companies that host them. It publishes an intelligence index built from its own test suite next to measured speed (output tokens per second), latency and cost per task, for proprietary and open-weight models, so you can build a shortlist on more than one number.
v4.3.2, 10 evaluations
artificialanalysis.ai · Oct 2026
Platform
AI & Machine Learning
Artificial Analysis
15 minutes per check, then quarterly
Intermediate
Free
What Artificial Analysis is
Artificial Analysis is an independent benchmarking site for AI models and the providers that serve them. Where a blind-vote leaderboard tells you which answers people prefer, Artificial Analysis runs its own fixed set of evaluations and measures performance directly, then puts quality, speed and price side by side.
What it measures
As of October 2026 the site tracks four main things:
- Intelligence Index: a combined score from ten evaluations (version 4.3.2 lists tests such as Terminal-Bench, SciCode, Humanity's Last Exam and long-context reasoning). It leans towards agentic work, coding and reasoning on economically useful tasks.
- Speed: output tokens per second, measured on real endpoints.
- Latency: how long a task takes to decode, reported separately from time to first token.
- Cost: the average cost to run one task from the index, broken down by token type, which is closer to what you pay than a price per million tokens.
It covers proprietary models, open-weight models and open models with commercial restrictions, and it compares different hosts of the same open model, which is useful because the same model can be much faster or cheaper on one provider than another.
How to use it well
Use it at the shortlist stage. Filter to models that fit your privacy and budget rules, look at the intelligence score against cost per task (the charts make the trade-off visible), and check speed for the host you would actually use. Then take three candidates and test them on your own data. Return every quarter: new releases and price changes show up here quickly.
Limits to keep in mind
The index measures the site's chosen tasks, in English, with its own prompts. It cannot tell you how a model handles your documents, your language mix or your tool list. Small score differences are within noise. Treat it as a well-measured starting point, not a verdict.
Free or paid?
The public leaderboards and charts are free to browse.
Best for: developers building a model shortlist, comparing hosts of open models, or doing a quarterly review of prices and new releases.
Ready to start?
Opens on Artificial Analysis in a new tab.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.