Which Ollama Local Model is Best? Sunflower SVG Comparison (16 Models)
· 34 mins read
Which Ollama Local Model is Best? Sunflower SVG Comparison (16 Models)
After testing LLMs on ducks, vehicles, marine life, chess, sports, and machines, we wanted to know: can today’s top models capture the mathematical beauty of nature? This time we asked our Ollama local models to draw a sunflower with seeds – a prompt that tests whether models understand Fibonacci spiral patterns (real sunflower seeds are arranged in intersecting clockwise and counter-clockwise spirals that follow Fibonacci numbers like 21, 34, 55, and 89), radial symmetry, color gradients (dark center seeds, bright yellow petals, green stem and leaves), and organic curves that blend mathematics with aesthetics.
The prompt was: Make an svg image of a sunflower with seeds
Why a sunflower? This prompt is a mathematical stress test for SVG generation because it combines: (1) Fibonacci spirals – real sunflower seeds follow Fibonacci sequences, and a model with strong mathematical intuition should attempt spiral arrangements rather than a simple grid of dots, (2) Two-scale structure – a sunflower has a dark seed-filled center and bright outer petals, requiring the model to distinguish two visual zones, (3) Radial symmetry – petals and seeds radiate from the center, testing the model’s understanding of rotational geometry and <transform> usage, (4) Color theory – yellow petals, brown or black seeds, green stem and leaves – the model must choose a coherent palette, (5) Organic vs. geometric – the model must balance mathematical precision (spirals) with organic softness (petal shapes), and (6) Scale and density – a real sunflower has hundreds of seeds; the model must decide how many to draw and how to arrange them without making the SVG file enormous.
The goal is not to declare a winner – it is to give you the data so you can pick the best model for your own use case. We show you the SVG, the stats, and a short analysis for each. You decide.
How to Choose the Best Ollama Model for Sunflower SVGs
The sunflower prompt rewards different things than previous prompts. Here are the criteria to use:
Fibonacci spiral pattern: Does the model arrange seeds in spiral patterns (clockwise and counter-clockwise)? Or does it just place them in a grid, concentric circles, or random dots? True Fibonacci spirals are the gold standard.
Petal detail: Does it draw distinct petals around the center? How many petals? Real sunflowers have 34, 55, or 89 petals (Fibonacci numbers).
Two-zone structure: Is there a clear dark seed center and bright petal ring? Or is it just a flat yellow circle?
Color depth: Does it use gradients for the seeds (dark brown to black) and petals (yellow to orange)? Or flat fills only?
Radial symmetry: Is the flower radially symmetric? Are the petals evenly spaced around the center?
Stem and leaves: Does it include a green stem and leaves for context? Or just the flower head floating in a void?
SVG code quality: Does it use <defs>, <use>, and transforms to efficiently generate repeating seeds and petals? Good code structure is a sign of model competence.
How It Works
The script discovers all locally installed models via the Ollama API (/api/tags), then sends the identical prompt through the OpenAI-compatible endpoint (http://localhost:11434/v1/chat/completions). Each model’s response is parsed for an <svg>...</svg> block, and the extracted SVG is saved for rendering with minimal post-processing (adding width="100%" height="auto" for responsive embedding and fixing XML errors so the SVG renders in browsers).
Unlike our cloud model benchmarks, these models run entirely on the local GPU – no cloud subscription or network round-trip required. This means generation times reflect local hardware performance, and model sizes range from 1B to 31B parameters. Embedding, vision, and OCR models are automatically skipped.
Summary Table: Compare All Models at a Glance
Use this table to quickly compare models on the metrics that matter. The verdict column is a one-line summary to help you shortlist – but read the per-model sections below for the full picture before you decide.
16 out of 25 models produced a valid SVG. The 9 that failed either returned an error or did not include a valid <svg>...</svg> block in their response.
Quick Recommendation by Use Case
If you just want a shortcut, here is which model to pick based on what you care about:
You want the most mathematically accurate seed spirals: look for models whose SVG shows visible spiral patterns in the seed arrangement
You want the most visually beautiful sunflower: pick models labeled “Very high” complexity in the table above
You want a small, efficient SVG for web embedding: pick models with “Compact” verdict
You want accurate botanical detail (petals, stem, leaves): check the per-model analysis for stem and leaf presence
You want gradient-rich coloring: look for models that used <linearGradient> or <radialGradient>
You want to compare within a model family: pick deepseek-v4-pro vs deepseek-v4-flash, or glm-5.1 vs glm-5.2
You want a balance of detail and speed: pick models labeled “Balanced” or “Detailed”
Now read on for the full per-model breakdown and judge for yourself.
1. cieloforge/qwen2.5-14B-instruct-spec:latest
SVG size: 3050 characters Complexity: Medium Shape elements: 25 Distinct colors: 2 Raw response: 3340 characters Generation time: 148.9s
Analysis
This SVG contains approximately 25 shape elements and uses 2 distinct colors.
Notable SVG techniques used:
Uses <defs> blocks for reusable components (a sign of well-structured, editable SVG)
Includes animation elements – the model attempted motion
Uses Bezier curves (C, Q, A path commands) for smooth organic shapes
Contains XML comments, showing the model’s internal documentation of its work
This is a compact SVG (under 5,000 characters) that uses minimal markup to convey the scene. The model prioritized efficiency and simplicity over elaborate detail. View raw SVG source ```xml
