Gemini 4 Argon — Google's frontier model for coding and cyber defense
Gemini 4 Argon is the frontier model Google announced on September 30, 2026, built for long-horizon software engineering, enterprise knowledge work and cybersecurity defense. It is rolling out to vetted cyber defenders first and has no public API yet.
- 1M-token output
- Long-horizon coding
- Knowledge work
- Cybersecurity defense
- Long-video understanding
Gemini 4 Argon pricing
These are Google's published prices for Gemini 4 Argon. Gemini 4 Argon cannot be called through SeedRouter yet, so there is no SeedRouter rate for it.
| Item | Introductory | After introductory period | Notes |
|---|---|---|---|
| Input tokens | $2.00 / 1M | $4.00 / 1M | Prompt and context |
| Output tokens | $10.00 / 1M | $20.00 / 1M | Generated tokens, up to the 1M-token output limit |
| Cached input tokens | $0.10 / 1M | Not announced | 95% off the input price |
Prices in USD as published by Google on September 30, 2026. Source: Google, Gemini 4 Argon announcement
What is Gemini 4 Argon?
Gemini 4 Argon is Google's new frontier model, announced on September 30, 2026. Google built it to sustain deep reasoning across complex, long-horizon workflows, and positions it for real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense.
Gemini 4 Argon can write up to 1M tokens in a single response, up from the previous 64K, so the model can reason through hundreds of thousands of tokens before it answers. Google is releasing it in phases: trusted cyber defenders first, then paid API customers and Google AI Ultra subscribers.
Announced September 30, 2026 · Developer: Google DeepMind · Status: limited access through Google's Fairwind Program
Who can use Gemini 4 Argon today
Google is releasing Gemini 4 Argon in phases while it takes part in the U.S. government's voluntary process for pre-release model access.
Now: Fairwind Program partners
Trusted cyber defenders in Google's Fairwind Program have access, alongside Google's own teams. Partners may give access only to their internal security, incident response or penetration testing teams.
Next: paid API and AI Ultra
Google says the wider release will start with paid API customers and Google AI Ultra subscribers, before developers, enterprises and consumers more broadly. It has not given a date.
On SeedRouter: not yet
Gemini 4 Argon cannot be called through SeedRouter today. When it can, this page will show its model ID, our live rate and the API reference.
What Gemini 4 Argon is built for
The strengths Google highlights in its announcement, with the scores it published.
Long-horizon coding
Google reports 77.9% on DeepSWE v1.1, a benchmark of real-world, long-horizon software engineering tasks, and uses Argon agents for large codebase migrations.
Enterprise knowledge work
Argon leads the Vals Index at 68.9%, which weights finance, coding, legal and tax work by contribution to U.S. GDP, and scores 51.3% on Zapier's AutomationBench.
Cybersecurity defense
Google trained Argon to find, validate and patch software vulnerabilities on its own. It ties for first place on CWE-bench v1 with 68%.
1M-token output
One response can run to 1M tokens, which Google says gives the model room to reason deeply and solve hard problems in one pass.
Charts, documents and video
Google reports a state-of-the-art 91.7% on LVBench for long-video understanding, and highlights chart analysis and work across a series of documents.
Prompt-injection resistance
Google calls Argon its most resilient model yet against indirect prompt injection, and says it leads Gray Swan's Indirect Prompt Injection benchmark.
Gemini 4 Argon use cases Google has described
Examples from Google's own teams and early partners, as reported by Google.
Codebase migration
Argon agents are migrating C and C++ code to Rust across Google, from core libraries such as re2 up to the 800K+ lines of the Fuchsia Zircon kernel.
Performance engineering
In libgav1, Argon agents replaced 32K lines of SIMD code with safe Rust. The decoder runs 2.7x faster than the earlier Rust port with identical video output.
Infrastructure optimization
A team of Argon agents found and applied memory optimizations across Google's data centers, freeing over 300 TiB of memory once rolled out.
Vulnerability research
Wiz used Argon to uncover a critical vulnerability in healthcare software used by hospitals worldwide, a risk earlier frontier models had missed.
Finance and legal research
In Google's results, Argon leads Vals Finance Agent v2 with 65.4% and Harvey's Legal Agent Benchmark with 19.6%.
Algorithm design
Google's quantum researchers used Argon to optimize a subroutine's qubit and gate cost, beating the published baseline by 40% in minutes.
How to get Gemini 4 Argon access
- 01
Check the Fairwind Program
Google's Fairwind Program admits approved cyber defense organizations only. Other teams have no way to get access yet.
- 02
Watch for the paid API release
Google says paid Gemini API customers and Google AI Ultra subscribers come next. No date has been announced.
- 03
Get a SeedRouter key now
Create an account and an API key, and build on the text models SeedRouter already serves.
- 04
Check this page at launch
When Gemini 4 Argon is available on SeedRouter, this page will list its model ID, rate and request format.
Gemini 4 Argon API status
As of October 1, 2026, there is no public Gemini 4 Argon API.
No model ID yet
Google's public Gemini API model list does not include Gemini 4 Argon, and Google has not published an API model ID for it.
No published limits
Request parameters and the context window are not published. SeedRouter will not guess them: the API reference comes with the model.
Price already announced
Google has published the API price: $2 input and $10 output per 1M tokens at launch, then $4 and $20.
Gemini 4 Argon benchmarks
Google's published results against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5. Argon leads on 12 of the 18 benchmarks, ties on CWE-bench v1 and trails on 5.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index · knowledge work | 68.9% | 63.1% | 65.8% | 67.0% |
| AutomationBench · knowledge work | 51.3% | 41.4% | 31.4% | 42.5% |
| Vals Finance Agent v2 · knowledge work | 65.4% | 53.5% | 58.9% | 58.6% |
| Harvey's Legal Agent Benchmark · knowledge work | 19.6% | 5.4% | 6.7% | 3.8% |
| DeepSWE v1.1 · agentic coding | 77.9% | 74.1% | 67.4% | 74.2% |
| FrontierSWE v2 · agentic coding | 55.0% | 65.5% | 56.3% | 62.3% |
| Vibe Code Bench · agentic coding | 91.9% | 89.6% | 90.3% | 90.3% |
| Terminal-bench 4.0 · agentic coding | 57.4% | 58.2% | 57.9% | 66.4% |
| PostTrainBench · ML engineering | 45.3% | 44.3% | 40.2% | 49.3% |
| Terminal-Bench Science 0.1 · science | 57.6% | 68.1% | 52.6% | 63.3% |
| LABBench 2 · science | 88.8% | 85.4% | 68.6% | 73.1% |
| RiemannBench · math | 76.0% | 72.0% | 65.6% | 69.6% |
| GraphWalks up to 128K · long context | 99.7% | 98.7% | 91.4% | 90.6% |
| GraphWalks 256K to 1M · long context | 84.2% | 71.8% | 65.0% | 66.8% |
| Agent's Last Exam · computer use | 39.5% | 34.2% | — | 38.2% |
| OSWorld-2.0 offline subset · computer use | 69.2% | 72.6% | — | — |
| Chartography · multimodal | 71.6% | 71.0% | 46.2% | 66.3% |
| LVBench · long video | 91.7% | 87.5% | 79.7% | 83.7% |
| CWE-bench v1 · cybersecurity | 68.0% | 68.0% | 58.0% | 67.0% |
Scores as published by Google on September 30, 2026, with Argon at its highest thinking setting. Results for other models are mostly their developers' own reported numbers; SeedRouter has not run these benchmarks. Source: Google DeepMind, Gemini 4 Argon
What is not known about Gemini 4 Argon yet
No general access
Only Fairwind partners and Google's own teams can use Gemini 4 Argon. Google has not set a date for paid API customers.
Context window not published
Google states the 1M-token output limit but not the input context window. Its long-context results include inputs of up to 1M tokens.
Introductory period length
Google has not said how long the $2 / $10 price lasts before $4 / $20 applies.
Benchmarks are Google's
The table above is Google's own. Argon trails GPT-6 Astra or Claude Opus 5.5 on FrontierSWE v2, Terminal-bench 4.0, PostTrainBench, Terminal-Bench Science and OSWorld-2.0.
Cyber guardrails depend on who you are
Google is releasing Argon without cyber guardrails only to trusted defenders and its own teams. Elsewhere it is designed to refuse harmful cyber and CBRN requests.
Why build on SeedRouter now
One key for many models
Call GPT-6, Claude, Kimi and DeepSeek text models, plus image and video models, with one SeedRouter API key.
Pay per use
Top up once and spend it on any model. No plan, no monthly fee.
Failures are free
A request that errors is not charged.
Live prices
Every model page shows the rate your account pays, read from billing.
Models you can call today
Text models available on SeedRouter now, at live rates.
Call GPT-6 Astra with the OpenAI SDK. OpenAI's most capable model, with a 1.05M-token context, live per-token pricing and a playground you can try right now.
View pricingCall GPT-6 Sol with the OpenAI SDK. 1.05M-token context, reasoning effort from none to max, live per-token pricing and a playground you can try right now.
View pricingCall GPT-6 Luna with the OpenAI SDK. The lowest-cost GPT-6 model, with a 1.05M-token context, live per-token pricing and a playground you can try right now.
View pricingTry Claude Sonnet 5.5 for code reviews and document questions. See live token prices, test a prompt, and check verified capabilities before connecting your app.
View pricingCall Claude Fable 5.1 with the Anthropic or OpenAI SDK. Deep reasoning, 1M-token context, cheap cached reads, live pricing and a playground to try it today.
View pricingCall Claude Fable 5 with the Anthropic or OpenAI SDK. Deep reasoning, 1M-token context, forced tool use, live per-token pricing and a playground to test it.
View pricingGemini 4 Argon guides
Release coverage and analysis for this model.
Gemini 4 Argon FAQ
Is Gemini 4 Argon out?+
Gemini 4 Argon was announced on September 30, 2026, but it is not generally available. It is rolling out first to trusted cyber defenders in Google's Fairwind Program; developers and consumers do not have access yet.
When will Gemini 4 Argon be available?+
Google has not given a date. It says the wider release will start with paid API customers and Google AI Ultra subscribers, after it has strengthened safeguards with feedback from early testers.
How much does Gemini 4 Argon cost?+
Google's introductory API price for Gemini 4 Argon is $2 per 1M input tokens and $10 per 1M output tokens, with cached input 95% off. After the introductory period it is $4 input and $20 output per 1M tokens.
What is Gemini 4 Argon?+
Gemini 4 Argon is Google's new frontier model, built for long-horizon software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. It can write up to 1M tokens in one response.
How does Gemini 4 Argon compare with GPT-6 Astra and Claude Opus 5.5?+
In the 18 benchmarks Google published, Gemini 4 Argon leads on 12, ties GPT-6 Astra on CWE-bench v1 and trails on 5, including FrontierSWE v2 and Terminal-bench 4.0. These are Google's own results.
Is there a Gemini 4 Argon API?+
Not publicly. Google has announced the API price but has not added Gemini 4 Argon to its public Gemini API model list or published a model ID.
Can I use Gemini 4 Argon on SeedRouter?+
Not yet. When Gemini 4 Argon can be called through SeedRouter, this page will show its model ID, our live rate and the API reference.
What is Gemini 4 Argon's output limit?+
Gemini 4 Argon can generate up to 1M tokens in a single response, up from the previous 64K. Google has not published its input context window.
What is Google's Fairwind Program?+
It is Google DeepMind's program for approved cyber defense organizations. Partners get exclusive access to Gemini 4 Argon, on its own or with the CodeMender security agent, to find and fix vulnerabilities.
Who makes Gemini 4 Argon?+
Gemini 4 Argon is made by Google. Google DeepMind develops the Gemini models.
Get ready for Gemini 4 Argon
Create a SeedRouter account and build on the models available today.