Google's new AI coder surprised a writer who had doubted it

A tech writer who has long found Google's Gemini disappointing at coding says its newest model, Gemini 4 Argon, did far more than it was asked. Mahnoor Faisal of MakeUseOf tested it against Anthropic's Claude Opus 5.5 by asking each to build a note-taking app.

What the test showed

Google announced Argon on September 30 and called it one of the best models for complex software engineering. The writer was doubtful. Earlier Gemini models often did the bare minimum, such as building a portfolio site with seven of more than 400 articles. This time the prompt was vague on purpose, to see how well each tool filled in the gaps. Argon finished its first build about 20 minutes before Claude Code. It also built the app around details the writer shared about being a computer science student and tech journalist. Unrequested extras included:

• An infinite canvas for brainstorming
• A knowledge graph that links notes together
• Executable code blocks for programming
• Spaced-repetition flashcards for studying
• A fact-checking ledger for claims and sources

What is not known yet

The part of the article available stops before the writer actually used the app. Until then, the feature list came from Argon's own report of its work. Whether the app runs well, and how it compares with Claude's version, is not yet clear. Anyone paying for AI to build software should check the finished product before trusting it.