Two models compared on the same three tasks, down to time and cost
This is a comparison that gave exactly the same three prompts to two of the latest models and recorded the outputs, the time taken and the actual charges side by side. Building an open-world game took 85 minutes versus 55, reconstructing ten photos into an editable 3D space with the Blender MCP took 2 hours 54 minutes versus 67 minutes, and making a 45-second clip with code alone took 68 minutes versus 18; on the same task there really are cases, such as $76 versus $27, where quality and cost flip at the same time. The most striking moment is when a user who did not know how to work the 3D tool simply put in a request and the model operated the tool directly through MCP to finish the job — a working example of the idea of hooking up in-house GUI tools and handing work over. However, all three tasks are creative works, the judging criteria are subjective impressions such as fun or a premium feel, and each task was run only once, so it is hard to stretch this into a general ranking of the models.
Key points
- Choose a model not by benchmarks but by a table of time, cost and pass/fail on 3–5 of your own tasks
- On the same task, quality and cost can flip together, as in $76 versus $27
- The model operating an external 3D tool by itself via MCP to produce output is the most transferable evidence
- Each task was run once and judged subjectively, so the sample is narrow
This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.