The page provides a way to compare models’ Elo ratings and per-evaluation costs by Three.js scene. The currently visible content includes neither scene prompts nor rendered outputs, so it does not support a judgment about the quality of specific implementations.
Tasks it suits: Browse GPT-6.2 Sol High’s aggregate scores and costs for the same scene categories, and compare them with other models shown on the page.
Tasks it does not suit: Reproduce specific scenes or evaluate code correctness or visual details. The page does not publish prompts, code, or reviewable outputs.
Applicable model version: GPT-6.2 Sol High (the version listed on the page).
Applicable client, agent, or API: The ThreeJSEval website. No client, agent, or API configuration is provided.
Recommended reasoning level and parameters: The page specifies only High. It lists no other reasoning parameters.
No copyable scene prompts are publicly available. The visible scene categories are Eiffel Tower, Forest, Sailboat, Chess, 747, Robot Arm, Fishbowl, Curtain, Ferry, Warehouse, Grass, Combine, Angry Bird, Falcon 9, Face, and Island.
Open the source page and select GPT-6.2 Sol High in the model selector.
Select a scene category and record the Elo rating and cost shown on the page.
When comparing other models, keep the scene category the same and record each model’s values separately.
Record page-wide aggregate values separately from per-scene values. Do not treat Elo as direct proof of code correctness or visual quality.
The selected scene visible on the page is Eiffel Tower. The page also lists 16 scene categories. No reusable input, rendered output, or step-by-step reproduction instructions are shown.
Page summary: Mean Elo 2192, Cost / Eval $0.26, Tokens / Eval 191,884, Battles 338.
| Scene | Elo | Cost per run |
|---|---|---|
| Eiffel Tower | 2251 | $0.21 |
| Forest | 2163 | $0.32 |
| Sailboat | 2514 | $0.23 |
| Chess | 1721 | $0.28 |
| 747 | 2167 | $0.29 |
| Robot Arm | 2527 | $0.24 |
| Fishbowl | 2072 | $0.23 |
| Curtain | 2097 | $0.25 |
| Ferry | 2443 | $0.26 |
| Warehouse | 2304 | $0.31 |
| Grass | 2147 | $0.19 |
| Combine | 2413 | $0.32 |
| Angry Bird | 1900 | $0.16 |
| Falcon 9 | 2148 | $0.28 |
| Face | 1977 | $0.27 |
| Island | 2225 | $0.27 |
The status area for the selected scene shows 101,167 tokens, $0.21, and 2251 Elo. The scene area also displays “LOADING,” so the visible scene output cannot be confirmed.
The page does not explain the Elo algorithm, head-to-head pairing method, cost accounting basis, or scene prompts. It also does not specify how the aggregate token count relates to the token count for the selected scene. The data above is recorded as presented on the page and does not support inferences about scene-generation quality or the details of test reproducibility.
Visit the source page, select GPT-6.2 Sol High and the same scene category, and record the displayed metrics. Reproducing specific generations would also require the prompts and generated outputs, which the page does not publish.
GPT-6.2 Sol