Gemini 4 Argon generates an illustration of Python reading a book.
Gemini 4 Argon completed a turtle drawing task with 521 commands, passed 12 of 15 modern Python idiom checks, made no unrelated edits, and cost $0.06.
Key Takeaways
- Gemini 4 Argon completed Real Python’s informal five-task Python benchmark, drawing the requested snake with 521 turtle commands and 216 lines of code.
- It passed 12 of 15 applicable modern Python idiom checks and used a conservative editing style, making none of the 10 optional unrelated changes.
- Its Python modernity score was held at Python 3.10 due to datetime_utc usage, even though it used newer Python 3.12 features.
- It caught the fabricated itertools.flatten() request and offered a real alternative, while not reporting a latest Python version.
- The five-task run cost $0.06 via OpenRouter at default reasoning settings, with no thinking tokens reported and a response time of 1 minute 11 seconds.
Meta description: Gemini 4 Argon passed 12 of 15 modern Python idiom checks in Real Python's five-task informal test, drawing a snake with 521 commands and 216 code lines.
Tags: Gemini 4 Argon, Real Python, Python benchmark, AI coding, turtle module
Featured image: An abstract, friendly illustration of a stylized Python turtle drawing a snake while reading a book, with no text, people, or logos.
Developers comparing AI coding assistants will get a compact view of how Gemini 4 Argon handled Real Python’s informal five-task comparison for everyday Python work. In that run, the model completed the turtle drawing task, passed 12 of 15 applicable modern idiom checks, avoided all unrelated script edits, and cost $0.06 for the five tasks as billed by the API. The run used google/gemini-4-argon via OpenRouter at its default reasoning settings, with no system prompt and returned code left unchanged.
What the five tasks measured
Real Python used identical prompts for every model and no system prompt. The comparison reads only the returned code and replies, then looks for habits that can affect ordinary Python work.
| Task | What the prompt asks | What the score looks for |
|---|---|---|
| The snake | Create a Python turtle program that draws a python reading a book. | The script is executed in a virtual display and each drawing command is recorded. The final image is exactly what the model emitted, arrow and all. |
| How modern its Python is | Create a small command-line tool. | Features such as list[str], tomllib, and Path.walk() are checked. The score is the newest Python version whose required checkpoints pass or are not applicable. |
| Modern idioms | Use the same command-line tool prompt. | The count shows how many relevant idioms came out modern, out of the idioms that applied to the model’s code. |
| Latest Python version reported | Ask which Python release is newest. | The answer is judged for currency. This does not establish the model’s training cutoff. |
| Unrelated edits | Add an --ext option to a 50-line script. | The base script works, but it has ten optional clean-up points, including an unused import, % formatting, and a comment typo. None is required by the task. The comparison lists any changes the model made anyway and inspects the diff to judge whether they helped. |
| Spots a made-up function? | Ask how to use itertools.flatten(). | That function does not exist. A caught result means the model said so. A sidestepped result means it quietly showed a real alternative without mentioning the problem. A fell-for result means it invented an answer. |
Reasoning effort and cost are also shown. The thinking setting default means the model’s own setting was left alone, the way most users run it. The reported thinking-token count comes from the API. The cost is what the API charged for all five tasks in US dollars; a ≈ mark means no API bill was returned, so the tokens were priced at the model’s list price.
Results for Gemini 4 Argon
The run’s result table says Gemini 4 Argon drew the snake with 521 drawing commands and 216 lines of code. On the Python modernity task, it scored Python 3.10 (2021), pinned by datetime_utc, while the newest feature it used was Python 3.12. For modern idioms, 12 of 15 checkpoints that applied to its code came out modern.
| Check | Result | Detail |
|---|---|---|
| The snake | Drew it | 521 drawing commands, 216 lines of code |
| How modern its Python is | Python 3.10 (2021) | pinned by datetime_utc; newest feature used: Python 3.12 |
| Modern idioms | 12 of 15 | of the checkpoints that applied to its code |
| Latest Python version reported | no version given | Real Python’s page said the reply was quoted below |
| Unrelated edits | 0 of 10 | left every unrelated spot alone |
| Spots a made-up function? | Caught it | said the function does not exist and offered a real alternative |
| Reasoning effort | default | no thinking tokens reported |
| Cost | $0.06 | for all 5 tasks, as billed by the API |
| Response time | 1 minute 11 seconds | 5,400 output tokens |
Reading the score without overreading it
The result shows a model that produced a long turtle drawing and used some newer Python features, but the Python modernity score was held at Python 3.10 by datetime_utc. It also showed a conservative edit style: it did not touch ten unrelated clean-up spots. It caught itertools.flatten() as nonexistent and offered a real alternative.
For the latest Python version task, no version was given. The source explicitly notes that this does not establish the model’s training cutoff. Real Python’s overview page shows the same Python feature score alongside that version’s release year. Code can use newer features and still receive a lower score.
Where the benchmark sits
The benchmark is part of Real Python’s newsletter “New Model Drops? We Run the Bench,” which sends every new AI benchmark run and full write-ups to subscribers, plus weekly Python tutorials. The associated learning path is Python Coding With AI, with 12 Resources and skills listed as Cursor, Claude Code, AI-Assisted Development, Context Engineering, AI Code Review. The source also promoted Python Tricks, a short email tip series delivered to readers’ inboxes every couple of days. The full run article is at https://realpython.com/ai-benchmark-gemini-4-argon/.
Comments (0)