The author says that, with the same prompt and extra-high reasoning, Gemini 4 Hydrogen took 15 minutes and cost $1.20, but needed a second version to produce a usable result. Sonnet 5.5 took 32 minutes and cost $2.60, and the author found its 3D design more creative. Because the prompt, runtime environment, number of iterations, and basis for calculating costs were not disclosed, this is a personal case and cannot establish typical speed or value for money.
What this comparison can help assess: The differences in reported time, price, and usability in the author's one 3D coding comparison.
What should not be inferred: The models' speed, creativity, or overall value for money on other 3D tasks, coding tasks, clients, or reasoning settings.
Model versions covered: Gemini 4 Hydrogen and Sonnet 5.5; specific model IDs and revision numbers were not provided.
Test environment or client: Not specified.
Reasoning tier and parameters: The author says both used the same prompt and extra-high reasoning; the platform, parameters, and precise meaning of that tier were not specified.
The author gave both models the same prompt for a 3D coding task and reported elapsed time, price, and a subjective assessment of creativity. Hydrogen needed a V2 to produce a usable result. The original post did not disclose the prompt, task specifications, code or finished output, model invocation method, tools and harness, first-version results, per-iteration time and costs, or whether Sonnet was iterated.
Sonnet 5.5: 32 minutes, $2.60.
Gemini 4 Hydrogen: 15 minutes, $1.20; the author says it needed a V2 to produce a usable result.
Based on the figures reported in the original post, Hydrogen's reported time was about 47% of Sonnet's (15 ÷ 32), and its cost was about 46% (1.20 ÷ 2.60). However, the post does not say whether Hydrogen's 15 minutes and $1.20 include V2, so the full cost of reaching a usable result cannot be compared.
The author subjectively rated Sonnet 5.5 as more creative at 3D coding; the post provided no scoring criteria or blind evaluation.
| Model | Time reported in original post | Cost reported in original post | Usable result and author's assessment |
|---|---|---|---|
| Sonnet 5.5 | 32 minutes | $2.60 | The author found it more creative at 3D coding; whether it was iterated was not specified |
| Gemini 4 Hydrogen | 15 minutes | $1.20 | Needed a V2 for a usable result; whether the cost includes V2 was not specified |
This case reports speed, cost, and information about one iteration: Hydrogen's first-pass or aggregate figures appear faster and cheaper, but the author explicitly says it needed a V2 to be usable. Since it is unclear whether the reported cost and time include V2, $1.20 and 15 minutes cannot be treated as the total cost of reaching a usable result. The creativity judgment reflects the author's personal impression, not an independent evaluation.
The full original post is visible on X. After checking the visible replies, no further details from the author about this test were found. Although the post includes a video, it does not provide the prompt, run logs, or complete settings needed to reproduce the test. The conclusion applies only to this personal test, whose details were not fully disclosed.
Exact reproduction is not possible: the post does not disclose the prompt, task input, model endpoint, client or agent, tool configuration, outputs from each version, or time and cost broken down by iteration. A comparable test should publish these settings and record, for each run, the time and cost from the first attempt until the result meets a predefined usability standard. It should also specify the reasoning-tier settings used for each model.
Gemini 4 Hydrogen