- Published on
Gemini 4 Argon Goes GA β 1M Output Tokens and a New Frontier for Agentic Coding
From 64K to 1 million output tokens. That single change captures how big a leap Gemini 4 Argon represents over its predecessors.
Google made Gemini 4 Argon generally available on October 13, 2026 β two weeks after a limited rollout through its Fairwind Program for cyber defenders. On launch day, access opened across the Gemini app, Gemini API, AI Studio, Google Antigravity, and Search''s AI Mode.
Built for Agentic Work
Google describes Argon as "our most capable model built for complex coding, agentic, math, and science tasks." The emphasis on agentic tasks is the telling part.
Where earlier models were optimized for what they know, Argon is optimized for what it can do. In agent mode it writes code, runs it, reads the output, and decides the next step on its own.
Google''s internal data: teams using Argon inside Antigravity saw agentic code submissions rise more than 35% while code rollbacks fell by nearly a third. The model also handles large-scale C/C++ to Rust migrations autonomously.
Key Numbers
| Metric | Gemini 4 Argon | Notes |
|---|---|---|
| Max output tokens | 1,000,000 | Up from 64K |
| Context window | 1M tokens | β |
| DeepSWE v1.1 | 77.9% | vs GPT-6 Astra 74.1%, Claude Opus 5.5 74.2% |
| FrontierSWE v2 | 55.0% | GPT-6 Astra leads at 65.5% |
| Terminal-bench 4.0 | 57.4% | Claude Opus 5.5 leads at 74.2% |
A million output tokens is not just a number. It means generating an entire novel, processing a massive codebase in one context, or running an agent task over many hours without losing its working memory.
Where It Trails
Argon does not lead every benchmark. On FrontierSWE v2 and Terminal-bench 4.0, GPT-6 Astra and Claude Opus 5.5 respectively come out ahead. As always with AI benchmarks, the best model depends on the task at hand.
Access and Pricing
- Gemini app + Search AI Mode: Google AI Pro and AI Ultra subscribers
- Antigravity: AI Ultra plan required
- Gemini API / AI Studio: Paid API customers
Reported API pricing: 10 per million output tokens, with a planned increase to 20 starting December. Verify official pricing with Google before committing.
Why It Matters for Education
The 1-million-token output window matters for EdTech in a specific way: long-horizon tutoring and curriculum agents. An agent that can hold a full semester''s worth of student work in its context can give coherent, cumulative feedback across essays or projects without losing earlier context. That was hard before; it is practical now.
Source: