Google Opens Gemini 4 Argon API to Developers with 1M-Token Output Limit
Google has expanded access to its Gemini 4 Argon frontier model, making it available today to paid API developers, Google AI Ultra subscribers, and Pro tier users. Announced initially on September 30, 2026, with a restricted rollout for cybersecurity defenders, Argon introduces a 1 million token output limit and aggressive introductory pricing of $2 per million input tokens, positioning it as a direct competitor to OpenAI’s GPT-6 Astra and Anthropic’s Claude Opus 5.5.
How Gemini 4 Argon Benchmarks and Pricing Compare to Rivals
Google reports that Argon leads in 13 of 19 benchmark rows against competing frontier models. In software engineering, the model scores 77.9% on DeepSWE v1.1, outperforming Claude Opus 5.5 (74.2%) and GPT-6 Astra (74.1%).
| Model | DeepSWE v1.1 Score | Gray Swan Prompt Injection (ASR) | Max Output Tokens |
|---|---|---|---|
| Gemini 4 Argon | 77.9% | 0.7% | 1,000,000 |
| Claude Opus 5.5 | 74.2% | Not published | Not published |
| GPT-6 Astra | 74.1% | Not published | Not published |
In cybersecurity evaluations, Argon achieves a 0.7% attack success rate on Gray Swan’s Indirect Prompt Injection benchmark, the lowest among 13 tested models. Despite these frontier capabilities, Google is pricing Argon at roughly a fifth of the cost of OpenAI’s GPT-6 Astra and Anthropic’s Fable 5 models.
Why Google Prioritized Cybersecurity in Gemini 4 Argon
Unlike previous frontier releases that launched directly to the public, Google initially restricted Argon to the Fairwind Program, a security initiative for vetted cyber defenders. The model was explicitly trained for defensive cyber work, enabling it to autonomously find, validate, and patch critical software vulnerabilities.
Tulsee Doshi, Google’s Gemini model product lead, stated that "starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible".
This phased approach allowed Google to scale safeguards in areas like prompt injection before broader deployment. Internally, the model is already executing complex engineering tasks. According to Lorphic, Argon agents recently replaced “32,000 lines of hand-written SIMD code with safer”, auto-vectorized Rust, producing a memory-safe video decoder that runs 2.7 times faster. Google also notes the model is halving the computing power needed for YouTube’s comments section and debugging the drone physics simulator for Wing.
What the 1 Million Token Output Limit Means for Gemini 4 Argon Developers
The most significant architectural shift in Argon is its output capacity. Previous Gemini models capped at 64,000 output tokens, but Argon raises this ceiling to 1 million tokens in a single generation pass.
For enterprise developers and IT professionals, this 16x increase fundamentally alters how agentic workflows are constructed. When a model can generate across hundreds of thousands of tokens in one trajectory, it sustains long tasks—such as migrating massive legacy codebases or analyzing extensive legal discovery documents—without being forced to stop, summarize, and re-prompt. This reduces the need for complex state-management and chunking logic in multi-step agent pipelines, allowing developers to hand off deeper work and get back finished outputs in a single pass. Google notes the model is "built to sustain deep reasoning across complex, long-horizon workflows", changing how the company builds software.
How Enterprises Can Access Gemini 4 Argon via Google Cloud
Developers can begin building with Argon using the paid tier of the Gemini API. Google AI Ultra subscribers can also test the model in Google AI Studio and the Antigravity agent-first development platform.
The introductory API pricing is set at “$2 per million input tokens and $10 per million output tokens”, which will eventually rise to standard rates of $4 and $20. For enterprise teams operating on Google Cloud, Argon is now integrated to support complex software builds, financial analysis, and cybersecurity deployments like CodeMender.
While Argon sets new records in real-world software engineering and cybersecurity, its true test will be how effectively enterprises can integrate a 1 million-token output model into production environments without exceeding latency or compute budgets.
Related articles
