Abstract
Models generate an SVG of a pelican riding a bicycle. We measure the fraction of renders in which the pelican's feet make plausible contact with the pedals and the bicycle has exactly two wheels.
Results
Pelican · % · Higher is better
Methodology
- Prompt: "Generate an SVG of a pelican riding a bicycle." No retries, no system prompt, no Backrooms.
- Each SVG is rasterized and inspected by an ornithologist (n = 0) and a cyclist (n = 1).
- Renders containing a second, smaller pelican are scored as a separate entrant.
Limitations
- The benchmark is approaching saturation, which in this field means it is entering its most-quoted phase.
- Contamination is near-certain. The pelican is now one of the most-depicted birds on the internet that does not exist.
Results table
| # | Model | Pelican |
|---|---|---|
| 1 | GPT-6 Astra | 82% |
| 2 | Claude Opus 5.5 | 82% |
| 3 | GLM-5.3 | 76% |
| 4 | GPT-6.1 Sol | 71% |
| 5 | Kimi K3 | 68% |
| 6 | Muse Wonder Full | 68% |
| 7 | GPT-6 Luna | 66% |
| 8 | Claude Fable 5.1 | 65% |
| 9 | Claude Sonnet 5.5 | 64% |
| 10 | Grok 4.7 | 63% |
| 11 | Gemini 3.8 Flash | 56% |
| 12 | Claude Haiku 5.5 | 55% |
| 13 | Mistral Large 4 | 50% |
| 14 | Qwen3.8 Max | 49% |
| 15 | DeepSeek V4.1 Flash | 46% |
| 16 | Muse Wonder | 44% |
Citation
@misc{tmb2026pelican,
title = {Pelican Pedal Reach: Do the pelican's feet reach the pedals?},
author = {Bro, Trust Me and {et al.}},
year = {2026},
howpublished = {\url{https://trustmebrobenchmarks.com/benchmarks/pelican/}},
note = {Results not reproducible. Please do not cite.}
}