In this guide
Short answer
Google announced Gemini 4 Argon on October 1, 2026, but it did not make the model broadly available. The first cohort is a set of trusted cyber defenders in Google's Fairwind Program. Google says paid API customers and Google AI Ultra subscribers will follow after more guardrail work and feedback. Buyers should treat the announcement as an early release boundary, not a product they can safely budget or migrate to today.
Free AI tool buyer checklist
Check before you buy your next AI tool.
Get the weekly top 5 AI tools, one practical workflow and one way to reduce AI spending. Includes a free buyer checklist.
What Google actually announced
Argon targets long-horizon coding, enterprise knowledge work, and cybersecurity. Google lists an introductory future price of $2 per million input tokens and $10 per million output tokens, with cached input at a 95% discount. It also advertises an output ceiling of one million tokens. That ceiling is capacity, not a recommendation: long trajectories can increase latency, tool use, review work, and output cost even when they improve task completion.
Internal results need external reproduction
Google reports code migration, memory optimization, quantum-computing, and benchmark results from its own deployments and evaluations. These are useful hypotheses for buyer tests, not independent comparisons. The evidence needed next includes reproducible task sets, model and harness settings, failure distributions, reviewer effort, security escapes, completed-task cost, and results from organizations outside Google.
The safeguard story is part of the product
Google describes misuse defenses, prompt-injection testing, reasoning-based misalignment monitoring, red teaming, and hardened sandboxes. Those controls are especially important for a model designed to work across long trajectories and tools. Buyers should ask which controls ship in the API, which depend on Google's internal environment, what customers must operate themselves, and how false positives, missed attacks, logs, appeals, and incident response work.
Do not migrate on a launch post
Prepare a fixed evaluation set now: representative tasks, adversarial inputs, permission boundaries, acceptance rules, cost ceilings, rollback, and a baseline against the incumbent. Run it only when the exact API, region, terms, quotas, model identifier, data handling, and production support are documented. Compare accepted outcomes—not maximum tokens or a benchmark headline.
Bottom line
Argon may become a consequential frontier model, and its staged release is itself meaningful. Today, the prudent decision is to monitor access and pre-register a test. Availability, independent reproduction, control behavior, and full task economics still need proof.
Transparency
How this guide was checked
About this article
How this article was prepared
- Evaluation type
- Source-based research
- Last reviewed
- Evidence
- 4 listed sources
We label hands-on testing separately. Research-based articles draw on product documentation and other cited sources, rather than direct use of every plan. Read the full methodology.
Editorial basis
Sources and testing
- Editorial basis
- Source-based research
- Primary references
- 4
- Products covered
- 1
- Last checked
- 2026-10-02
Important limits
- • The model is not broadly available, so DiscoverAI could not run an independent production evaluation.
- • Performance, safety, savings, and benchmark figures are Google-reported and may depend on internal tools, data, infrastructure, and review processes.
Sources and references
References cited by this article. Check the original source for scope, dates and current terms.
Frequently asked questions
Is Gemini 4 Argon publicly available?
Not broadly. Google says it is initially rolling out to trusted cyber defenders, with paid API customers and Google AI Ultra subscribers intended to follow.
How much will Gemini 4 Argon cost?
Google announced introductory pricing of $2 per million input tokens and $10 per million output tokens, plus a cached-input discount. Final access and commercial terms still need verification.
Does a one-million-token output limit make tasks cheaper?
No. It increases headroom. Longer trajectories can increase token charges, latency, tool calls, review work, and failure exposure; measure total cost per accepted task.
Should a team migrate to Argon now?
No migration decision is possible until the exact production service, terms, controls, regional availability, quotas, and model behavior can be tested against representative work.
Read next

