Benchmarks & Leaderboards
A benchmark measures a defined task. Read the methodology and date, and validate the result against your own use case.
An editorial shortlist for this task. PICK marks our editorial starting point. The order is not a universal benchmark ranking.
-
Artificial AnalysisCompare model quality, speed and cost
-
ArenaCompare models with human preferences
-
SWE-benchEvaluate code agents on repository tasks
-
Epoch AIResearch on AI progress and capabilities
Observed benchmark snapshot · 13 November 2026
| Benchmark | Model | Metric |
|---|---|---|
| Intelligence Index ↗ | Claude Fable 5.2 (max) | 65 |
| AA-Image-T2I v2.1 ↗ | GPT Image 2.6 Sunburst (max) | Elo 1261 ±14 |
| AA-Video-T2V v2.1 ↗ | Wan 3.0 | Elo 1175 ±13 |
These models headed the observed tables. Video uncertainty intervals overlap. The three metrics measure different tasks and cannot be compared with each other. Open the sources for current standings.
How to make a good choice
A benchmark measures a defined task. Read the methodology and date, and validate the result against your own use case. Try a representative example with the same input in two or three tools. Consider quality, output control, export options and total cost. Always check the current product page and terms.
Which tool is best?
That depends on your task and requirements. The shortlist above is a starting point. A benchmark for one model or task cannot fully describe the quality of an entire app.
Are these tools free?
Providers set their own prices, trials and limits. Open the official link to check the current terms. This directory does not make undated pricing promises.
Source check · 13 November 2026 · Selection methodology