gpt-5.6-sol
Official Codex
All evaluations
Folkbench pelican gallery
Same prompt, what pelican and bicycle will each model invent?
This page collects animations from official APIs and relays. Each work keeps its station, group, and model so the result can be viewed as generated.
This is a playful comparison, not a verdict. Differences on this page do not measure model quality or relay quality, and they are not used in Folkbench public scores or rankings.
Shared promptFirst generationRaw resultHTML + SVG
02 / PUBLISHED RESULTS
Evidence from this evaluation.
Only released snapshots with a named window, sample, and method version belong here.
Gallery available
This is a playful comparison, not a verdict. Differences on this page do not measure model quality or relay quality, and they are not used in Folkbench public scores or rankings.
Open the result galleryGPT series
Claude series
claude-opus-5
anpin.ai · 高缓cc_kiro
Hunyuan series
hy4-preview
Official Hunyuan
Gemini series
gemini-3.5-flash
Official Gemini
GLM series
glm-5.3
Official GLM
Other models
kimi-k3
Official Kimi
deepseek-v4-pro
Official DeepSeek V4 Pro
deepseek-v4.1-flash
Official DeepSeek V4.1 Flash
03 / RELATED WRITING
Read the thinking behind the test.
Articles explain the task design, review decisions, and what the evidence can or cannot support.
04 / RUN IT YOURSELF
Make one authorized observation of your own.
A user run is private and bounded. It does not create a public score or change a ranking.
05 / EVIDENCE LIMITS
What this test cannot prove.
A single suite is one signal. It does not certify a model, station, security posture, or future availability by itself.