The author says GPT-6.1 Terra completed this 3D helicopter task quickly but rates the result's quality very poorly; this is a subjective account of one case with no full methodology disclosed.
Suitable for assessing: The author's impressions of the speed and quality of this 3D helicopter generation result.
Not suitable for extrapolating to: Terra's general 3D capabilities, performance on other tasks, or a cross-model quality ranking.
Model version covered: The original post names GPT-6.1 Terra.
Test environment or client: Not specified.
Reasoning tier and parameters: Not specified.
The author posted a video of a 3D helicopter case and gave a subjective assessment of its speed and quality. In a reply, the author said the test used a very strong prompt and that the prompt was not the problem. However, the full prompt was not made public, and no execution records that could be independently checked were provided. It is therefore impossible to independently confirm whether Terra and the other models mentioned by the author used exactly the same input.
Speed: The author subjectively considered generation fast; no actual runtime was provided.
Quality: The author considered the result poor and regarded this experience as a failure.
Cross-model information: Replies mention examples involving Opus 5.5, Gemini 4 Carbon, Qwen 4, and Kimi K3.1, but the post provides no data sufficient to verify input consistency or make a comparison.
Sample size: The post reports one Terra case; it does not say whether the task was run more than once.
Test prompt: The full prompt was not made public.
Test process and harness: Not specified.
Runtime: Not specified; the author gave only a qualitative assessment.
Reasoning tier and parameters: Not specified.
Cost: Not specified.
Quantitative quality metrics or scores: None provided.
This post records one author's negative assessment of a single 3D helicopter generation experience, describing it as fast but poor in quality. The prompt, environment, number of runs, and quality criteria are all missing; the author's claims about the prompt's strength and that the prompt was not the problem cannot be independently verified. This case can therefore serve only as personal feedback: it does not represent the model's overall capability or support cross-model conclusions.
The original post does not provide the full prompt, client, execution parameters, or run records, so the test cannot be reproduced under the original conditions. To verify it, preserve the complete input and output, and record the model, client, reasoning tier, parameters, number of runs, runtime, and cost. A cross-model comparison would also require fixed inputs and execution conditions.
GPT-6.1 Terra
