Text to Three.js / CASE 02 / 2026-08-18
One Pi Harness, four models: how does each build an Airbus H145?
Using exactly the same Prompt in Pi Harness, Grok 4.5, Grok 4.6, Kimi K3 and GLM 5.2 each generate an interactive Three.js helicopter HTML.
- Harness
- Pi Harness
- Prompt
- Identical
- Variable
- Model
- Outputs
- 4 Three.js HTML files
EXPERIMENT SETUP / 01
Same Pi Harness Prompt
All four runs use Pi Harness and change only the model. Each model has to build, in a single HTML file, a 3D Airbus H145 model that can be rotated, zoomed and inspected.
create a 3D model of an Airbus H145 helicopter in Three js in htmlAbove is the complete Prompt shared by all four runs, with the original capitalization and wording preserved.
RESULTS / 02
Four models, four Three.js outputs
Each result is a standalone raw HTML that can be used full screen. Once it is open you can drag to rotate the view, scroll to zoom, and directly compare fuselage structure, rotor detail, materials and scene rendering.
COMPARISON / 03
Cost comparison under the same Harness
This run recorded the task cost of each model. Run time and human scores were not provided, so they are explicitly marked as not recorded and no subjective ranking is added.
| Harness | Model | Time | Cost | Score | Conclusion |
|---|---|---|---|---|---|
| Pi Harness | Grok 4.5 | Not recorded | $0.10 | — | See the full result |
| Pi Harness | Grok 4.6 | Not recorded | $0.10 | — | See the full result |
| Pi Harness | Kimi K3 | Not recorded | $0.53 | — | See the full result |
| Pi Harness | GLM 5.2 | Not recorded | $0.19 | — | See the full result |
VERDICT / 04
Raw results kept, no subjective ranking for now
For now this Case keeps the shared Prompt, the four runnable results and the real costs, and does not declare a winner on the reader's behalf without an evaluation. Later, scores for structural accuracy, fidelity to H145 features, visual quality and interaction completeness can be added from the same viewing angle.
KEEP EXPLORING