Same prompt, same settings
Every model answers the same prompt with the same resolution, length and frame. We re-encode each result, strip its metadata and trim clips to one length, so nothing but the picture gives it away.
Blind model arena
Two anonymous results from one prompt. Pick the one you would keep; the names appear after you vote.
Image and videoOnly logged-in votes countRatings refresh hourlyLeaderboard →How it works →
Your verdict
Prompt
A
B
Keyboard: A · B · T for a tie · X if both are weak · F for full screen
Current standings
Full leaderboard →Bradley–Terry ratings, ranked by rating. The 95% interval shows how settled each one is.
Image
Updated Oct 19, 2026
Current top image models in the SpicyAPI arena: GPT Image 2.5 Flare (1202), GPT Image 2.5 Sunburst (1187), and GPT Image 2 (1144). 37 image models are ranked as of October 19, 2026.
| Rank | Model | Rating | Votes | |
|---|---|---|---|---|
| 1 | GPT Image 2.5 Flare OpenAI | 1202 | 95% interval 1034–1353 | fewer than 50 votes |
| 2 | GPT Image 2.5 Sunburst OpenAI | 1187 | 95% interval 1059–1334 | fewer than 50 votes |
| 3 | GPT Image 2 OpenAI | 1144 | 95% interval 1005–1290 | fewer than 50 votes |
| 4 | Nano Banana 2 | 1124 | 95% interval 988–1283 | fewer than 50 votes |
| 5 | Qwen Image 3.0 Alibaba | 1071 | 95% interval 959–1187 | fewer than 50 votes |
| 6 | Jib Mix Qwen J1B | 1045 | 95% interval 898–1215 | fewer than 50 votes |
| 7–8 | Qwen Image 3.0 Pro Alibaba | 1039 | 95% interval 885–1123 | fewer than 50 votes |
| 7–8 | Nano Banana Pro | 1039 | 95% interval 879–1169 | fewer than 50 votes |
| 9 | Krea 2 Large Krea | 1038 | 95% interval 940–1157 | fewer than 50 votes |
| 10 | Seedream 5.0 Lite ByteDance | 1028 | 95% interval 928–1137 | fewer than 50 votes |
Video
Updated Oct 19, 2026
Current top video models in the SpicyAPI arena: Seedance 2.0 (1094), Wan 3.0 (1091), and Seedance 2.0 Fast (1089). 35 video models are ranked as of October 19, 2026.
| Rank | Model | Rating | Votes | |
|---|---|---|---|---|
| 1 | Seedance 2.0 ByteDance | 1094 | 95% interval 993–1251 | fewer than 50 votes |
| 2 | Wan 3.0 Alibaba | 1091 | 95% interval 981–1188 | fewer than 50 votes |
| 3 | Seedance 2.0 Fast ByteDance | 1089 | 95% interval 972–1213 | fewer than 50 votes |
| 4 | Seedance 2.0 Mini ByteDance | 1087 | 95% interval 981–1191 | fewer than 50 votes |
| 5 | FLUX 3 Black Forest Labs | 1085 | 95% interval 889–1267 | fewer than 50 votes |
| 6 | Wan 3.0 Pro Alibaba | 1072 | 95% interval 947–1217 | fewer than 50 votes |
| 7 | Seedance 2.5 ByteDance | 1070 | 95% interval 960–1181 | fewer than 50 votes |
| 8 | Wan 3.0 Prime Alibaba | 1069 | 95% interval 941–1169 | fewer than 50 votes |
| 9 | MiniMax H3 Max MiniMax | 1055 | 95% interval 945–1164 | fewer than 50 votes |
| 10 | MiniMax H3 MiniMax | 1053 | 95% interval 974–1158 | fewer than 50 votes |
Waiting for first votes (9)
| Rank | Model | Rating | 95% interval | Votes |
|---|---|---|---|---|
| — | Seedance 2.5 SpicyByteDance | — | — | — |
| — | Wan 3.0 SpicyAlibaba | — | — | — |
| — | MiniMax H3 SpicyMiniMax | — | — | — |
| — | Wan 3.0 Prime SpicyAlibaba | — | — | — |
| — | Vidu Q4 PreviewVidu | — | — | — |
| — | MiniMax H3 Singularity LoRAMiniMax | — | — | — |
| — | Seedance 2.0 SpicyByteDance | — | — | — |
| — | Wan 2.6 SpicyAlibaba | — | — | — |
| — | HunyuanVideo 1.5Tencent | — | — | — |
How the arena works
Names stay hidden until you vote
The battle you see carries no model name, file name or ID. Once your vote is in, both models are revealed and that pair is closed.
Which votes count
Anyone can vote. Only logged-in votes shape the ratings; anonymous votes are recorded separately and never appear in the counts. Votes cast too fast to have looked, bursts from one account or network, and accounts that always pick the same side are kept but not counted.
How the rating works
We fit a Bradley–Terry model to all counted votes every hour and show it on the familiar Elo scale with a 95% interval. A tie and “both are weak” each count as half a win for both sides.
Who runs it
We sell every model in the arena. Prices and margins play no part in which pairs you see or how they are scored, and every result is reviewed before it enters the pool. Some prompts are moody or glamorous, but nothing in the pool goes further than a fashion campaign would.