Ollama Cloud Models SVG Comparison: A Duck Jumping From a Plane

Following up on our previous duck-on-a-bicycle comparison, we sent a new, more challenging prompt to every cloud model in the local Ollama registry. This time the prompt was: Make an svg image where a duck is jumping from a plane and landing with a parachute.

This prompt tests the models on a more complex scene: a dynamic action (jumping), two distinct objects (duck and plane), spatial relationships (falling from sky), and a specific mechanism (parachute). It is a harder test of scene composition than the bicycle prompt.

How It Works

The script discovers all cloud-hosted models via the Ollama API (/api/tags), pulls each model, 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.

Cloud models are identified by the remote_host field in the API response – these models are hosted on Ollama Cloud rather than running locally. This means even very large models (671B parameters) can be queried instantly without local GPU resources.

Summary of Results

#	Model	SVG Size (chars)	Status
1	deepseek-v4-pro_cloud	8308	Generated
2	gemma4_31b-cloud	2795	Generated
3	glm-5.1_cloud	15528	Generated
4	glm-5.2_cloud	11128	Generated
5	kimi-k2.6_cloud	7141	Generated
6	minimax-m2.7_cloud	3089	Generated
7	nemotron-3-super_cloud	1606	Generated
8	nemotron-3-ultra_cloud	15952	Generated
9	qwen3.5_397b-cloud	3277	Generated
10	bjoernb/claude-opus-4-5:latest	-	Retired (410)
11	deepseek-v3.1:671b-cloud	-	Retired (410)
12	glm-5:cloud	-	Retired (410)
13	qwen3-vl:235b-cloud	-	Retired (410)

9 out of 13 models successfully produced a valid SVG. The 4 retired models returned HTTP 410 Gone (they were removed from Ollama Cloud on 2026-07-15).

1. deepseek-v4-pro_cloud

SVG size: 8308 characters
Raw response: 8682 characters

Analysis

This SVG contains approximately 77 shape elements and uses 21 distinct colors.

Notable SVG techniques used:

• Uses <defs> blocks for reusable components (a sign of well-structured SVG)
• Uses <use> references to reuse defined elements, reducing duplication
• Includes gradient fills for richer visual depth
• Uses transforms (translate, rotate, scale) for precise positioning
• Contains XML comments, showing the model’s internal documentation of its work

This is a moderately detailed SVG (5,000-10,000 characters), balancing visual richness with code clarity. The model included enough detail to make the scene recognizable while keeping the markup manageable.

View raw SVG source

```xml ```

2. gemma4_31b-cloud

SVG size: 2795 characters
Raw response: 3807 characters

Analysis

This SVG contains approximately 21 shape elements and uses 6 distinct colors.

Notable SVG techniques used:

• Uses transforms (translate, rotate, scale) for precise positioning
• Adds <text> labels, showing the model tried to annotate the scene
• 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 ```

3. glm-5.1_cloud

SVG size: 15528 characters
Raw response: 23137 characters

Analysis

This SVG contains approximately 119 shape elements and uses 60 distinct colors.

Notable SVG techniques used:

• Uses <defs> blocks for reusable components (a sign of well-structured SVG)
• Includes gradient fills for richer visual depth
• Applies SVG filters (e.g., drop shadows) for visual effects
• Uses transforms (translate, rotate, scale) for precise positioning
• Adds <text> labels, showing the model tried to annotate the scene
• Contains XML comments, showing the model’s internal documentation of its work

With over 10,000 characters of SVG markup, this is one of the most detailed outputs in the comparison. The model invested significant effort in adding fine details, the plane fuselage, parachute cords, and sky scene elements.

View raw SVG source

```xml ```

4. glm-5.2_cloud

SVG size: 11128 characters
Raw response: 13180 characters

Analysis

This SVG contains approximately 83 shape elements and uses 26 distinct colors.

Notable SVG techniques used:

• Uses <defs> blocks for reusable components (a sign of well-structured SVG)
• Uses <use> references to reuse defined elements, reducing duplication
• Includes gradient fills for richer visual depth
• Uses transforms (translate, rotate, scale) for precise positioning
• Includes animation elements for motion effects
• Contains XML comments, showing the model’s internal documentation of its work

With over 10,000 characters of SVG markup, this is one of the most detailed outputs in the comparison. The model invested significant effort in adding fine details, the plane fuselage, parachute cords, and sky scene elements.

View raw SVG source

```xml ```

5. kimi-k2.6_cloud

SVG size: 7141 characters
Raw response: 13972 characters

Analysis

This SVG contains approximately 23 shape elements and uses 19 distinct colors.

Notable SVG techniques used:

• Uses <defs> blocks for reusable components (a sign of well-structured SVG)
• Includes gradient fills for richer visual depth
• Applies SVG filters (e.g., drop shadows) for visual effects
• Uses transforms (translate, rotate, scale) for precise positioning
• Includes animation elements for motion effects
• Contains XML comments, showing the model’s internal documentation of its work

This is a moderately detailed SVG (5,000-10,000 characters), balancing visual richness with code clarity. The model included enough detail to make the scene recognizable while keeping the markup manageable.

View raw SVG source

```xml ```

6. minimax-m2.7_cloud

SVG size: 3089 characters
Raw response: 3295 characters

Analysis

This SVG contains approximately 34 shape elements and uses 12 distinct colors.

Notable SVG techniques used:

• Uses transforms (translate, rotate, scale) for precise positioning
• Includes animation elements for motion effects
• 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 ```

7. nemotron-3-super_cloud

SVG size: 1606 characters
Raw response: 1879 characters

Analysis

This SVG contains approximately 16 shape elements and uses 8 distinct colors.

Notable SVG techniques used:

• Uses transforms (translate, rotate, scale) for precise positioning
• 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 ```

8. nemotron-3-ultra_cloud

SVG size: 15952 characters
Raw response: 17298 characters

Analysis

This SVG contains approximately 82 shape elements and uses 34 distinct colors.

Notable SVG techniques used:

• Uses <defs> blocks for reusable components (a sign of well-structured SVG)
• Includes gradient fills for richer visual depth
• Applies SVG filters (e.g., drop shadows) for visual effects
• Uses transforms (translate, rotate, scale) for precise positioning
• Includes animation elements for motion effects
• Adds <text> labels, showing the model tried to annotate the scene
• Contains XML comments, showing the model’s internal documentation of its work

With over 10,000 characters of SVG markup, this is one of the most detailed outputs in the comparison. The model invested significant effort in adding fine details, the plane fuselage, parachute cords, and sky scene elements.

View raw SVG source

```xml ```

9. qwen3.5_397b-cloud

SVG size: 3277 characters
Raw response: 3378 characters

Analysis

This SVG contains approximately 33 shape elements and uses 15 distinct colors.

Notable SVG techniques used:

• Uses transforms (translate, rotate, scale) for precise positioning
• 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 ```

Retired Models

The following models were listed in the Ollama registry but returned HTTP 410 Gone when queried. They were retired from Ollama Cloud and are no longer available for inference:

• bjoernb/claude-opus-4-5:latest – glm-4.7 was retired at 2026-07-15
• deepseek-v3.1:671b-cloud – deepseek-v3.1:671b was retired at 2026-07-15
• glm-5:cloud – glm-5 was retired at 2026-07-15
• qwen3-vl:235b-cloud – qwen3-vl:235b was retired at 2026-06-16

Comparison With the Bicycle Prompt

Compared to the previous duck-on-a-bicycle comparison, this parachute prompt is more challenging because it requires:

• Two distinct objects: a duck and a plane (vs. just a duck + bicycle)
• Vertical composition: sky, falling motion, ground (vs. horizontal ground scene)
• A specific mechanism: parachute with cords and canopy (vs. just two wheels)
• Spatial relationships: duck below plane, parachute above duck

Observations across models:

• Most models successfully rendered the parachute as a canopy shape (often a semicircle or arc)
• Several models added motion lines to indicate falling, which was not present in the bicycle scene
• The plane was the most varied element: some drew a simple rectangle, others a detailed fuselage with windows
• Sky backgrounds (clouds, sun) were more common in this prompt than ground backgrounds in the bicycle prompt

Conclusion

This comparison shows that 9 out of 13 active Ollama cloud models can generate valid SVG artwork from a complex natural language prompt involving multiple objects, spatial relationships, and a specific mechanism. The results vary dramatically in complexity, style, and technique:

• Simplest output: nemotron-3-super:cloud produced a compact SVG with basic shapes
• Most detailed output: nemotron-3-ultra:cloud generated a 15,000+ character SVG with rich scene elements
• Most technically advanced: deepseek-v4-pro:cloud used <defs>, <use>, and transforms

All SVGs were generated with zero post-processing – what you see is exactly what each model produced. The variety of approaches (from minimalist geometric to richly detailed scenes) highlights how different training data and model architectures lead to distinctly different creative outputs, especially when the prompt demands complex scene composition.

Links

• Previous: Duck Driving a Bicycle Comparison
• Ollama Official Website
• Ollama Cloud Documentation
• SVG Specification (MDN)

Enjoyed this post? Never miss out on future posts by following us