Updated 4 November 2026 · Artificial Analysis
Which AI should I use?
Answer three questions and get one model to use, plus two close alternatives, picked from independent benchmark data on what each model scores, costs and how fast it writes.
The short answers
| If you want | Use |
|---|---|
| The smartest model overall | Gemini 4 CarbonGoogle |
| The fastest good model | GPT-6.1 TerraOpenAI |
| The cheapest good model | Claude Haiku 5.5Anthropic |
| Running it yourself (open weights) | Kimi K3.1 ProKimi |
| Reading very long documents | GPT-6 AstraOpenAI |
How the quiz picks
Every answer is a plain rule, applied to the same numbers as the rankings. Nothing is weighted behind the scenes.
Smartest answers: highest Artificial Analysis Intelligence Index. For coding, highest Coding Index instead, so models that haven’t been scored on coding are left out.
Speed: the shortest time to a complete 500-token answer (wait for the first token, then write) among models scoring at least 36 on the Intelligence Index, 60% of the top score, so a fast but weak model never wins.
Lowest cost: lowest blended price per 1M tokens among the same models scoring 36+.
Running it myself: the smartest model whose weights are published, so you can download and run it.
Long documents: only models with a context window of at least 500K tokens (about 375,000 words), then ranked by the rule above.
Budget: drops every model whose blended price (3 parts input to 1 part output) is above $5 or $1 per 1M tokens. If nothing is left, the quiz tells you and ignores the budget.
Data: Artificial Analysis (intelligence and coding indexes, prices, speed) and OpenRouter (open weights, context windows). Updated 4 November 2026; refreshed every 6 hours.
More tools
AI cost calculator · Embed the leaderboard · All AI model rankings · Price table