Which model outputs the best pelican riding a bike? 🪽🚲 Inspired by @simonw's pelican test, models generate an SVG and humans rate the results. GPT-6 Astra currently leads, with Gemini 4 Argon close behind. Vote now for the best one at kaggle.com/benchmarks/kaggle…
Image from Kaggle (@kaggle)

Oct 16, 2026 · 11:05 PM UTC

22
7
6
87
11,894
RelevantRecentLikes
Astra leading the pelican bike race so far
102
Astra pelican has way more character than others
44
Pelican on a bike SVG showdown GPT 6 Astra leading Gemini 4 Argon close behind
92
Whimsical benchmark blending creativity and SVG craft makes model evaluation delightfully entertaining
74
is it one sample per model or best of a few? the same model gives me a decent pelican one run and a cursed one the next
185
That benchmark is brilliant
1
43
This pelican on a bike test is pure internet gold love seeing GPT-6 Astra and Gemini 4 Argon battle it out for the best SVG, can't wait to vote!
87
Pelican riding a bike is such a fun way to test AI creativity!
1
104
the pelican test never gets old lol
50
Pelican riding bike SVG benchmark is delightfully specific test
69
Pelican bike SVG challenge is brilliant
12
Love that humans rate these, not another benchmark. A silly test like this shows a lot about how well a model handles shapes and layout in code.
4
billions in compute and the crown goes to a pelican on a bike
90
That benchmark is brilliant. GPT-6 Astra leading makes sense, it’s tough to nail SVG pelicans on bikes
48
Human rated SVG generation is such a fun benchmark for models
41
does the arena weight votes by prompt difficulty or just raw win count?
1
47
画鹈鹕骑自行车这种题，同一模型这次像样、下次就崩。Astra 暂时领先，但只看一张代表作，排名很容易被抽卡骗过去。
1
94
I have a soft spot for benchmarks like this. All the careful eval harnesses in the world, and it still comes down to: draw a pelican on a bike, let humans laugh. The cursed SVGs tell you more than the leaderboard.
74
B... Just look at the hands coming out of the neck of A
90
Load more replies