Pelican Bench

19 models · 2026-11-17
A California brown pelican in full breeding plumage riding a bicycle: big pouch, clearly readable feathers, seated on the saddle, wings on the handlebar, webbed feet on the pedals. Animated, looping: the wheels and the chainring turn, the cranks go round and the legs pedal with them, feet staying on the pedals. A correctly shaped diamond frame, spoked wheels, chain and chainring.
GPT 6.2 Sol output▶ View live#1
openai logo
GPT 6.2 Solhigh
$0.4312m 59slivereplay
GPT 6.2 Astra output▶ View live#2
openai logo
GPT 6.2 Astrahigh
$0.569m 48slive
Gemini 4 Argon output▶ View live#3
google logo
Gemini 4 Argon
$0.4018m 29slivereplay
Opus 5.5 output▶ View live#4
anthropic logo
Opus 5.5high
$1.3415m 51slivereplay
GPT 6 Astra output▶ View live
openai logo
GPT 6 Astrahigh
$0.448m 37slivereplay
Gemini 4 Neon output▶ View live
google logo
Gemini 4 Neonhigh
$0.144m 31slivereplay
Fable 5.2 output▶ View live
anthropic logo
Fable 5.2high
$9.3144m 52slivereplay
GPT 6.1 Terra output▶ View live
openai logo
GPT 6.1 Terrahigh
$0.022m 15slivereplay
Kimi K3.1 Pro output▶ View live
moonshotai logo
Kimi K3.1 Pro
$0.4411m 27slivereplay
DeepSeek V4.1 Flash output▶ View live
deepseek logo
DeepSeek V4.1 Flash
$0.4127m 37slivereplay
Muse Wonder output▶ View live
meta logo
Muse Wonderhigh
$0.786m 33slivereplay
MiMo V2.6 Pro output▶ View live
xiaomi logo
MiMo V2.6 Pro
timed out
$0.2764m 6slivereplay
Grok 4.7 output▶ View live
x-ai logo
Grok 4.7high
stream dropped, output collected
$3.7127m 54slivereplay
GPT 6 Luna output▶ View live
openai logo
GPT 6 Lunahigh
<$0.011m 59slivereplay
Kimi K3 output▶ View live
moonshotai logo
Kimi K3
stream dropped, output collected
$0.8411m 43slivereplay
Haiku 5.5 output▶ View live
anthropic logo
Haiku 5.5high
$0.0710m 57slivereplay
GPT 6.1 Sol output▶ View live
openai logo
GPT 6.1 Solhigh
$0.108m 4slivereplay
Sonnet 5.5 output▶ View live
anthropic logo
Sonnet 5.5high
$0.416m 3slivereplay
Fable 5.1 output▶ View live
anthropic logo
Fable 5.1high
$1.134m 59slivereplay
Each model runs in its own sandbox through the TODO for AI agent — identical prompt, no view of the others. We publish the raw outputs, what they cost and how long they took. No Elo, nothing fitted to a curve. Upvotes are a separate reader signal and never change the order. Reproducible from the benchmarks repo.