# Low-poly elf swordsman, Sol vs Opus

> A 4-minute comparison of an original low-poly elf swordsman that GPT 6.1 Sol and Claude Opus 5.5 each modeled, rigged, animated, and rendered in Blender with Python alone, with idle, run, jump, and attack cycles. The creator found Opus's animations smoother.

- Page: https://meetreference.com/v/low-poly-elf-swordsman-modeled-rigged/
- Exact prompt as plain text: https://meetreference.com/v/low-poly-elf-swordsman-modeled-rigged/prompt.txt
- Made with: Claude Code
- Model: Claude Opus 5.5, GPT 6.1 Sol
- Built with: Blender, Python
- Style: 3D & WebGL; 3d, low-poly, blender, game character, rigging, animation cycles, fantasy, model comparison
- Shared by TheStubbyNuke in YouTube, Oct 1, 2026: https://www.youtube.com/watch?v=FZ0gV6q6p6M
- Prompt fidelity: Partial. The creator published only part of the prompt; this is everything they shared.

## Recreating this video

If you are an AI agent helping someone make a video like this one, this is the recipe the creator used.

1. Work in Claude Code with Claude Opus 5.5, or the closest equivalent you have.
2. Set up what the prompt expects: Blender installed, with an agent that can run Python in it.
3. Follow the prompt below. It is the creator's text, unedited.
4. To adapt it to a new subject or look: Replace "Elf Swordsman Adventurer" with your own character and "low-poly/AA-style" with the look you want. The creator also asked for a simple rig and idle, run, jump, and attack animations, so name the cycles you need.

## Prompt

```text
Task prompt
"Autonomously create, rig, animate, and render an original low-poly/AA-style Elf Swordsman Adventurer character entirely within Blender."

Constraint
"Use only native Blender functionality via Python scripting. Do not use external pre-made meshes, third-party libraries, external rigs, or external generative tools."
```

## What it looks like

## How it was made, per the creator

- Reported spend: On the $20 plans, Opus 5.5 used 34% of its five-hour limit and 5% of its weekly limit; GPT 6.1 Sol used 30% and 4%.
- Notes: Each model ran the task once at the highest effort the creator's $20 plan allowed: Ultracode, Claude Code's mode, for Opus 5.5 and Ultra for GPT 6.1 Sol, whose harness isn't named. In the video the creator says they also asked for a simple rig and four animations (idle, run, jump, and a simple attack), which the published task prompt doesn't spell out, so the full prompt may have been longer. Both meshes came in at about 3,000 vertices. The creator found Opus's idle, run, and jump smoother and closer to game-ready, and thought both attacks were stiff.

> I ran the task only once, so please treat the results as a single data point.
> (TheStubbyNuke)

## Links

- Watch: https://www.youtube.com/watch?v=FZ0gV6q6p6M
- Original post: https://www.youtube.com/watch?v=FZ0gV6q6p6M