Gemini 4 Argon vs Claude: How to Test It on Your Own Work
TL;DR
Google announced Gemini 4 Argon on September 30, 2026. It scores 77.9% on DeepSWE, a real-world coding test, but only trusted cyber defenders can use it today. Paid API customers and Google AI Ultra subscribers come next, with no date yet. If you build with Claude Code, don't switch. Run one real job on both models when Argon opens and keep the one that does it better.
👉 https://www.skool.com/claudecodeclub
What Gemini 4 Argon is
Gemini 4 Argon is Google DeepMind's new frontier model, announced on September 30, 2026. Google's post says it sets a new state of the art on DeepSWE v1.1 at 77.9%, a test of real-world, long coding tasks. Google also reports leading results on the Vals Index across finance, coding, legal and tax work. These are Google's own numbers.
Who can use Gemini 4 Argon today
Almost no one yet. Google is rolling Argon out to a set of trusted cyber defenders through its Fairwind Program. Paid API customers and Google AI Ultra subscribers come next. Google gave no date and says it wants to open Argon to developers, companies and everyday users as soon as possible.
What Gemini 4 Argon costs and what it can do
- Price: an intro rate of $2 per million input tokens and $10 per million output tokens, then $4 and $20. Google didn't say how long the intro rate lasts. Cached input is 95% off.
- Output size: up to 1 million tokens in one answer, up from 64 thousand.
- Google's own use: Argon agents are moving C and C++ code to the Rust language at Google, up to 800,000 lines for the Fuchsia Zircon kernel.
How to test Gemini 4 Argon against Claude
Google's post makes no comparison with Claude, and a benchmark score doesn't tell you how a model handles your work. So test it yourself. The rule: one real job, same instructions, both models. Pick something you already did with Claude Code that took real effort, like a bug fix, a new page, or a cleanup of a messy folder.
promptI want to compare two AI models on one real job. Help me pick the job: look at my recent work in this folder and suggest 3 tasks I already finished that took real effort. For each one, write the exact instructions I would give a new model so the test is fair.promptHere are the results from two models on the same job: [paste model 1 result] and [paste model 2 result]. Score each from 1 to 5 on: did it finish the job, did it break anything, how much fixing did I need to do, and how clear was the explanation. Tell me which one to use for this kind of job and why.What to do while you wait
- Keep building on Claude. Nothing about your setup has to change today.
- Write your test job down now. When Argon opens, you can run it in ten minutes.
- Watch the price. The $2 and $10 rate is an intro rate, so check the real rate before you plan around it.
- Trust your own test over any chart. Google's scores are Google's own.
👉 https://www.skool.com/claudecodeclub
Common questions
Can I use Gemini 4 Argon today?
Only if you're one of the trusted cyber defenders Google picked through its Fairwind Program. Paid API customers and Google AI Ultra subscribers are next, with no date given.
Is Gemini 4 Argon better than Claude?
Nobody outside Google can say yet. Google's post reports its own scores, such as 77.9% on DeepSWE, and makes no comparison with Claude. Test both on your own work.
How much does Gemini 4 Argon cost?
Google lists an intro price of $2 per million input tokens and $10 per million output tokens, then $4 and $20. Cached input is 95% off. Google didn't say how long the intro price lasts.
Should I switch from Claude Code?
No. Keep working as you are, and test Argon on one real job when it opens to you.
Keep going
Want the skills that make Claude Code worth every token?
Get 650+ plug-and-play skills, MCPs & prompts, plus 8,000+ members - $9/mo, cancel anytime.
Join the Club