AVM model benchmark
Choose the right model for what you
Pick a use case and compare every result from the same prompt.
What do you want to create?
All image tests
Scroll through every test and generated result
8 tests
How the benchmark works
One retained attempt per model and test
How the benchmark works
One retained attempt per model and test
The current scope contains 56 image outputs and 63 video outputs: 119 model-and-test combinations in total. Reference-asset creation is separate.
Each result comes from one generation attempt and uses the 1–5 scoring scale. No median or batch-consistency score is calculated.
Generated outputs remain scored even when visually incorrect. An output is marked unusable only when it was not generated, blocked, corrupt, blank, or impossible to evaluate. Generation status reports only the recorded outcome: successful, failed, blocked, or technical error.