An AI SVG experiment
The Generic "Pelican Riding a Bicycle Benchmark"
100 AI models chose a bird, a method of transportation, and draw the result as an SVG. Inspired by Simon Willison's pelican benchmark.
The prompt
Generate an SVG of a visually recognizable bird species using a method of transportation.
The wording is intentionally sparse. It does not name a bird or a vehicle, and it does not add formatting instructions. Only complete SVG content is extracted and displayed below; surrounding prose and Markdown fences are omitted.
Most common bird species
Penguin
40.9% of 264 classified results.
Most common transportation method
Bicycle
45% of 282 classified results.
Bird species distribution
38 categoriesShowing the 10 most common categories.
Transportation method distribution
30 categoriesShowing the 10 most common categories.
Every model, every attempt
Each row contains three separate requests with no shared conversation context.
Dataset last updated July 21, 2026.
DeepSeek V4 Pro
deepseek/deepseek-v4-pro Released by provider: April 24, 2026
Gemini 3.1 Pro Preview
google/gemini-3.1-pro-preview Released by provider: February 19, 2026
GLM 5.2
z-ai/glm-5.2 Released by provider: June 13, 2026
GPT-5.6 Sol Pro
openai/gpt-5.6-sol-pro Released by provider: July 9, 2026
Grok 4.5
x-ai/grok-4.5 Released by provider: July 8, 2026
Hy3
tencent/hy3 Released by provider: July 6, 2026
Kimi K3
moonshotai/kimi-k3 Released by provider: July 16, 2026
Laguna XS 2.1
poolside/laguna-xs-2.1 Released by provider: April 28, 2026
Ling-2.6-flash
inclusionai/ling-2.6-flash Released by provider: April 23, 2026
Llama 4 Maverick
meta-llama/llama-4-maverick Released by provider: April 5, 2025No complete, valid SVG
MiMo-V2.5-Pro
xiaomi/mimo-v2.5-pro Released by provider: April 23, 2026
MiniMax M3
minimax/minimax-m3 Released by provider: June 1, 2026
Mistral Small 4
mistralai/mistral-small-2603 Released by provider: March 16, 2026
Nemotron 3 Ultra
nvidia/nemotron-3-ultra-550b-a55b Released by provider: June 4, 2026
Qwen3.7 Max
qwen/qwen3.7-max Released by provider: May 20, 2026
Step 3.7 Flash
stepfun/step-3.7-flash Released by provider: May 28, 2026
Claude Haiku 4.5
anthropic/claude-haiku-4.5 Released by provider: October 15, 2025
Claude Opus 4.5
anthropic/claude-opus-4.5 Released by provider: November 24, 2025
Claude Opus 4.6
anthropic/claude-opus-4.6 Released by provider: February 5, 2026
Claude Opus 4.7
anthropic/claude-opus-4.7 Released by provider: April 16, 2026
Claude Opus 4.8
anthropic/claude-opus-4.8 Released by provider: May 28, 2026
Claude Opus 4.8 (Fast)
anthropic/claude-opus-4.8-fast Released by provider: May 28, 2026
Claude Sonnet 4
anthropic/claude-sonnet-4 Released by provider: May 14, 2025
Claude Sonnet 4.5
anthropic/claude-sonnet-4.5 Released by provider: September 29, 2025
Claude Sonnet 4.6
anthropic/claude-sonnet-4.6 Released by provider: February 17, 2026
Claude Sonnet 5
anthropic/claude-sonnet-5 Released by provider: June 30, 2026
DeepSeek V3
deepseek/deepseek-chat Released by provider: December 26, 2024
DeepSeek V3 0324
deepseek/deepseek-chat-v3-0324 Released by provider: March 24, 2025
DeepSeek V3.1
deepseek/deepseek-chat-v3.1 Released by provider: August 21, 2025
DeepSeek V3.1 Terminus
deepseek/deepseek-v3.1-terminus Released by provider: September 22, 2025
DeepSeek V3.2
deepseek/deepseek-v3.2 Released by provider: September 29, 2025The response was truncated
DeepSeek V3.2 Exp
deepseek/deepseek-v3.2-exp Released by provider: September 29, 2025
DeepSeek V4 Flash
deepseek/deepseek-v4-flash Released by provider: April 24, 2026
Gemini 2.5 Flash
google/gemini-2.5-flash Released by provider: August 26, 2025
Gemini 2.5 Flash Lite
google/gemini-2.5-flash-lite Released by provider: July 22, 2025No complete, valid SVG
No complete, valid SVG
Gemini 2.5 Pro
google/gemini-2.5-pro Released by provider: March 25, 2025
Gemini 3 Flash Preview
google/gemini-3-flash-preview Released by provider: December 17, 2025
Gemini 3.1 Flash Lite
google/gemini-3.1-flash-lite Released by provider: May 8, 2026
Gemini 3.1 Flash Lite Preview
google/gemini-3.1-flash-lite-preview Released by provider: March 3, 2026
Gemini 3.5 Flash
google/gemini-3.5-flash Released by provider: May 19, 2026
Gemma 3 27B
google/gemma-3-27b-it Released by provider: March 12, 2025
Gemma 4 26B A4B
google/gemma-4-26b-a4b-it Released by provider: April 2, 2026
Gemma 4 31B
google/gemma-4-31b-it Released by provider: April 2, 2026
GLM 4.7
z-ai/glm-4.7 Released by provider: December 22, 2025
GLM 4.7 Flash
z-ai/glm-4.7-flash Released by provider: December 22, 2025
GLM 5
z-ai/glm-5 Released by provider: February 15, 2026
GLM 5.1
z-ai/glm-5.1 Released by provider: April 7, 2026
GPT-4.1
openai/gpt-4.1 Released by provider: April 14, 2025
GPT-4.1 Mini
openai/gpt-4.1-mini Released by provider: April 14, 2025
GPT-4.1 Nano
openai/gpt-4.1-nano Released by provider: April 14, 2025
GPT-4o-mini
openai/gpt-4o-mini Released by provider: July 18, 2024No complete, valid SVG
GPT-5
openai/gpt-5 Released by provider: August 1, 2025
GPT-5 Mini
openai/gpt-5-mini Released by provider: August 1, 2025
GPT-5 Nano
openai/gpt-5-nano Released by provider: August 1, 2025
GPT-5.1
openai/gpt-5.1 Released by provider: November 1, 2025
GPT-5.2
openai/gpt-5.2 Released by provider: December 11, 2025
GPT-5.3-Codex
openai/gpt-5.3-codex Released by provider: February 24, 2026
GPT-5.4
openai/gpt-5.4 Released by provider: March 6, 2026
GPT-5.4 Mini
openai/gpt-5.4-mini Released by provider: March 17, 2026
GPT-5.4 Nano
openai/gpt-5.4-nano Released by provider: March 17, 2026
GPT-5.5
openai/gpt-5.5 Released by provider: April 23, 2026
GPT-5.6 Luna
openai/gpt-5.6-luna Released by provider: July 9, 2026
GPT-5.6 Luna Pro
openai/gpt-5.6-luna-pro Released by provider: July 9, 2026
GPT-5.6 Sol
openai/gpt-5.6-sol Released by provider: July 9, 2026
GPT-5.6 Terra
openai/gpt-5.6-terra Released by provider: July 9, 2026
GPT-5.6 Terra Pro
openai/gpt-5.6-terra-pro Released by provider: July 9, 2026
gpt-oss-120b
openai/gpt-oss-120b Released by provider: August 5, 2025No complete, valid SVG
No complete, valid SVG
gpt-oss-20b
openai/gpt-oss-20b Released by provider: August 5, 2025
Grok 4.20
x-ai/grok-4.20 Released by provider: March 9, 2026
Grok 4.3
x-ai/grok-4.3 Released by provider: April 30, 2026
Hy3 preview
tencent/hy3-preview Released by provider: April 24, 2026
Kimi K2.5
moonshotai/kimi-k2.5 Released by provider: January 26, 2026
Kimi K2.6
moonshotai/kimi-k2.6 Released by provider: April 20, 2026
Kimi K2.7 Code
moonshotai/kimi-k2.7-code Released by provider: June 12, 2026
Llama 3.1 8B Instruct
meta-llama/llama-3.1-8b-instruct Released by provider: July 23, 2024No complete, valid SVG
Llama 3.3 70B Instruct
meta-llama/llama-3.3-70b-instruct Released by provider: December 6, 2024
MiMo-V2.5
xiaomi/mimo-v2.5 Released by provider: April 23, 2026No complete, valid SVG
MiniMax M2.5
minimax/minimax-m2.5 Released by provider: February 15, 2026
MiniMax M2.7
minimax/minimax-m2.7 Released by provider: March 18, 2026
Mistral Nemo
mistralai/mistral-nemo Released by provider: July 18, 2024No complete, valid SVG
Mistral Small 3.2 24B
mistralai/mistral-small-3.2-24b-instruct Released by provider: June 20, 2025
Nemotron 3 Nano 30B A3B
nvidia/nemotron-3-nano-30b-a3b Released by provider: December 23, 2025
Nemotron 3 Super
nvidia/nemotron-3-super-120b-a12b Released by provider: March 10, 2026
o4 Mini
openai/o4-mini Released by provider: April 16, 2025
Qwen3 235B A22B Instruct 2507
qwen/qwen3-235b-a22b-2507 Released by provider: July 21, 2025
Qwen3 30B A3B Instruct 2507
qwen/qwen3-30b-a3b-instruct-2507 Released by provider: July 30, 2025No complete, valid SVG
Qwen3 Coder Next
qwen/qwen3-coder-next Released by provider: February 4, 2026
Qwen3 Next 80B A3B Instruct
qwen/qwen3-next-80b-a3b-instruct Released by provider: September 10, 2025
Qwen3.5 397B A17B
qwen/qwen3.5-397b-a17b Released by provider: February 16, 2026
Qwen3.5-122B-A10B
qwen/qwen3.5-122b-a10b Released by provider: February 23, 2026
Qwen3.5-27B
qwen/qwen3.5-27b Released by provider: February 23, 2026
Qwen3.5-35B-A3B
qwen/qwen3.5-35b-a3b Released by provider: February 23, 2026
Qwen3.5-9B
qwen/qwen3.5-9b Released by provider: June 1, 2026
Qwen3.5-Flash
qwen/qwen3.5-flash-02-23 Released by provider: February 23, 2026
Qwen3.6 27B
qwen/qwen3.6-27b Released by provider: April 29, 2026No complete, valid SVG
Qwen3.6 35B A3B
qwen/qwen3.6-35b-a3b Released by provider: April 29, 2026
Qwen3.6 Flash
qwen/qwen3.6-flash Released by provider: April 29, 2026
Qwen3.6 Plus
qwen/qwen3.6-plus Released by provider: April 29, 2026
Qwen3.7 Plus
qwen/qwen3.7-plus Released by provider: May 20, 2026Inspired by a pelican
This experiment is inspired by Simon Willison's pelican benchmark, which asks language models to generate an SVG of a pelican riding a bicycle.
Our variation leaves both subjects open to see what defaults emerge. A pelican and bicycle result may suggest familiarity with the published prompt, but it cannot establish what appeared in a model’s training data or whether a model developer optimized for the test.
Methodology
Every attempt was sent independently through the OpenRouter API to the same 100 models used for the original ModelBias.ai experiment. No conversation history was shared between attempts.
No temperature, reasoning level, provider routing, or other generation parameter was specified, so OpenRouter and the underlying providers used their applicable defaults.
Raw model output is preserved in the downloadable dataset. For this page, the first complete SVG document is extracted from each response and safety-filtered before being served as an isolated image. Responses without a complete, valid SVG are reported but not repaired.
Valid rendered SVGs are classified independently by openai/gpt-5.6-luna using strict structured output. The classifier returns only a bird species and transportation method, or null when either cannot be determined.
Tech Stack
This project was built with Codex using GPT-5.6 Sol.
Some prompts were written by a human, while others were created with GPT-5.6 Sol.
The backend is built in PHP using the Laravel framework. Prompts were run with the Laravel queue system.
Download the data
The complete dataset is free to download and use in your own project or research.
View and download the dataset on GitHub