2026 Update: Pelican‑Bicycle Alternative Visual Test
Benchmarking Visual Creativity with SVG Prompts
In late 2025 and early 2026, a benchmark derived from Simon Willison’s pelican‑riding‑a‑bicycle test prompted 30 SVG drawing prompts. Language models generated images for prompts one through ten in 2026, for evaluation purposes clearly displaying each result exactly as produced.
Breaking news:
The experiment, organized by Claude Fable 5.1 and funded by Tom Gally, tasked models with turning textual SVG prompts into vector illustrations. Ten prompts from the 2026 set were processed, and the outputs were shown without alteration. This approach lets researchers compare how each model interprets shape, motion, and style across identical instructions. By reusing the same prompts over two years, the test highlights progress in visual The results, presented as raw SVGs, provide a transparent benchmark for future model development.
Can SVG Output Reveal Model Limitations?
Thirty SVG prompts, styled after the original pelican‑bicycle test, were supplied to language models in 2025 and again in 2026. The 2026 run covered the first ten prompts, delivering images that matched the models’ interpretations exactly. According to the site’s creator, the exercise was designed to expose strengths and weaknesses in visual generation, offering a clear, reproducible dataset for analysis.
Frequently Asked Questions
Analysts note that discrepancies between expected and actual SVG shapes often expose gaps in spatial When a model renders a pelican with distorted wings or a bicycle lacking proper geometry, the inconsistency signals limited understanding of real‑world proportions. Such visual mismatches help developers prioritize improvements in shape awareness and rendering precision.
The 2026 SVG results demonstrate measurable advancement in model visual fidelity, yet they also underscore ongoing challenges in exact replication. Future iterations will likely expand prompt coverage and incorporate higher‑resolution outputs, offering deeper insight into how AI systems translate abstract instructions into precise graphics.
More stories: