Compare by task
See what each model produced from the same brief, then compare the reported result, cost and time.
Continuous shot
Orbit a stone horse statue once, in a single slow take, and finish where the shot started.
View test conditions
Conditions summarized from the original prompt.
- One slow orbit with no cuts
- Keep the statue centered
- Return to the starting view
-
Seedance 2.0 Fast
6/6$0.73165.0 s -
Grok Imagine Video 1.5
5/6$1.1273.4 s -
Grok Imagine Video
4/6$0.56107.1 s -
Grok Imagine Video 1.5 Lite
3/6$0.2444.0 s
Cost vs. result
Criteria passed
Cost per result (USD)
Minimal image edit
Replace a sign’s lettering while preserving the sign, wall, lighting and framing.
View test conditions
Conditions summarized from the original prompt.
- Change the lettering on the sign
- Preserve the wall and lighting
- Keep the original framing
-
K3 Edit
2/3$0.01012.3 s -
GPT Image 2
2/3$0.02020.8 s
Cost vs. result
Criteria passed
Cost per result (USD)
Habit tracker app screen
Generate a straight-on mobile mockup with legible small text and a familiar app layout.
View test conditions
Conditions summarized from the original prompt.
- A TODAY header
- Five habit rows with checkboxes and streak counts
- A bottom tab bar with four icons
-
FLUX.2 Pro
Pass$0.03010.4 s -
FLUX.2 Max
Pass$0.07030.8 s
Evaluation method
These are selected OpenRouter runs. Fractional scores show criteria passed on each task. The app mockup uses the source’s Pass label.
Confidence intervals are meaningful for published estimates that include them. These per-run task results have no published intervals.
How Arena reports uncertainty for aggregate scores ↗


