Gemini 4 Argon: What Long Output Means for Your Business

Google says Gemini 4 Argon can generate up to 1 million tokens in a single run, up from 64K. That makes long jobs, like code migration, possible without constant restarts. It is live now at $2 input and $10 output per million tokens until 30 November.

Key takeaways

Long output is Argon’s headline feature (Google’s claim)
Useful for big, multi-step jobs like code migration and bug fixing
Google reports 77.9% on DeepSWE v1.1; verify on your own code
Launch price ends 30 November, then it doubles

Why long output matters

Most AI tools stop after a few thousand words. You then ask them to continue, and they sometimes lose track of the earlier part. Google says Argon can keep going for up to 1 million tokens, which is roughly 750,000 words. In simple terms, one long job can run from start to finish.

One caution: independent testers measured a lower limit in their own set-ups. So test the limit on your own task.

Where it is live

Argon is available through the paid Gemini API (gemini-4-argon), Google AI Studio, Vertex AI and OpenRouter. In the Gemini app and AI Mode in Search, it is open to Google AI Pro and Ultra users. It is not on the free tier.

What Argon is built for

Codebase work. Google reports 77.9% on DeepSWE v1.1, a test of fixing real GitHub issues. It also says Argon agents are moving large C/C++ systems into memory-safe Rust, including an 800,000+ line kernel.
Safer agents. Google lists indirect prompt-injection defences as a pre-launch priority. Gray Swan testing shows a 0.7% attack success rate. This matters when an agent reads web pages or documents you don’t control.
Direct answers. Google describes Argon as giving clearer answers with less hedging.

A practical example for an SME

Say you run a software firm in Kolkata with a 10-year-old PHP billing module that nobody wants to touch. Do not hand over the whole system. Pick one module, ask Argon to rewrite one file at a time, write tests for each file, and have a developer review every change. [CONFIRM: add ZTS Infotech’s own test result here, only if the team has actually run one.] This keeps risk low and shows you the real cost per file.

Pricing

Launch pricing is $2 per million input tokens and $10 per million output tokens. Cached input is 95% cheaper. From 1 December, the rates double to $4 and $20. Because long output means more output tokens, calculate your cost at the December price too.

Frequently asked questions

What is Gemini 4 Argon’s biggest feature?

Long output. Google says it can generate up to 1 million tokens in one run, up from 64K.

Can Argon migrate old code?

Google says its agents are moving large C/C++ systems into Rust. For your own project, start with one module and review every step.

Is Argon free?

No. It is available through the paid Gemini API, and in the Gemini app for Google AI Pro and Ultra.

Are the benchmark scores independent?

Some are, some are Google-reported, including DeepSWE. Verify on your own workload.