What is Gemini 4 Argon?
Gemini 4 Argon is Google’s flagship reasoning model for long-running software engineering, enterprise knowledge work, science and defensive cybersecurity. Google first announced Argon on September 30, 2026 through a restricted Fairwind rollout, then made the standard model generally available on October 13, 2026.
Developers can now call Argon through the paid Gemini API in Google AI Studio and Google Cloud under the model ID gemini-4-argon. Google AI Pro and Ultra subscribers can use it in the Gemini app, while Ultra subscribers also receive access through Google Antigravity. A separate cyber-permissive version remains restricted to approved Fairwind projects.
QUICK VERDICT: Gemini 4 Argon is now a real production option rather than an announced model waiting for access. It combines a shared one-million-token input, thinking and output window with strong provider-reported results in coding, enterprise work and long-context reasoning. Its introductory API price is competitive, but it doubles after November 30, 2026, and long-running outputs still require strict cost, latency and safety controls.
Field | Verified value |
|---|---|
Provider | |
Initial announcement | September 30, 2026 |
General availability | October 13, 2026 |
License | Proprietary |
API model ID | gemini-4-argon |
Context window | Up to 1 million tokens shared across input, thinking and output |
Inputs | Text, images, audio and video |
Output | Text |
Architecture | Sparse mixture-of-experts transformer |
Knowledge cutoff | March 2026 |
Introductory API pricing | $2 input and $10 output per million tokens through November 30, 2026 |
Standard API pricing | $4 input and $20 output per million tokens after November 30, 2026 |
Cached input | 95% below the applicable standard input-token price |
Best fit | Long-horizon coding, enterprise analysis, science and defensive cybersecurity |
Last verified | October 15, 2026 |
What changed from Gemini 3.8 Flash
Argon is a new frontier-model generation rather than a small point update. It is designed for deeper reasoning, longer task horizons, stronger multimodal analysis and complex work in coding, legal, finance, science and cybersecurity.
Area | Gemini 4 Argon change |
|---|---|
Generation | First generally available model in the Gemini 4 family |
Context design | Input, thinking and output share a unified window of up to one million tokens |
Task horizon | Designed for sustained, multi-step work rather than short single-turn answers |
Coding | Targets repository-scale migrations, optimization and agentic engineering |
Enterprise work | Built for legal, finance, tax, document and chart analysis |
Multimodality | Accepts text, image, audio and video inputs |
Availability | Generally available through the paid API, Gemini app and Google Cloud |
The practical question is whether Argon improves accepted-task quality, speed and total cost in the exact workflow. Keep prompts, tools, source material and scoring criteria fixed when comparing it with Gemini 3.8 Flash.
Capabilities and best use cases
Google positions Argon for work that must continue across many steps, files or tool calls. Its model card highlights deep investigations, engineering, multimodal document analysis, scientific and mathematical reasoning, and defensive cybersecurity. These are intended-use claims, not a guarantee that every workflow will complete reliably.
Capability | Practical use |
|---|---|
Software engineering | Large migrations, debugging, performance optimization and algorithm design |
Knowledge work | Legal, financial, tax and other document-heavy professional workflows |
Science and math | Technical research, quantitative analysis and theorem-oriented work |
Visual and multimodal understanding | Charts, long videos, audio and multi-document synthesis |
Cyber defense | Vulnerability discovery, remediation and defensive research |
Long-context work | Large source collections and extended reasoning within a shared one-million-token window |
Agent workflows | Long-running tasks with tools, checkpoints and iterative verification |
The strongest documented fit is long-running software engineering, enterprise knowledge work, science and defensive cybersecurity. A capable base model can still fail when retrieval is weak, tools are over-permissioned, instructions conflict or the system lacks verification. Evaluate the full application rather than the model in isolation.
Benchmark evidence
Google’s October model card reports strong results across enterprise work, coding, long context, multimodal understanding and cybersecurity. Argon leads the compared models on the Vals Index, AutomationBench, Vals Finance Agent v2, Harvey’s Legal Agent Benchmark, DeepSWE v1.1 and several multimodal and long-context tests. It does not lead every evaluation: GPT-6 Astra is higher on FrontierSWE v2 and Terminal-Bench Science, while Claude Opus 5.5 is higher on Terminal-bench 4.0 and PostTrainBench.
Evaluation | Gemini 4 Argon | Best comparison context |
|---|---|---|
Vals Index | 68.9% | Highest of the four models in Google’s table |
AutomationBench | 51.3% | Highest of the four models in Google’s table |
Harvey Legal Agent Benchmark | 19.6% | Highest of the four models in Google’s table |
DeepSWE v1.1 | 77.9% | Highest of the four models in Google’s table |
FrontierSWE v2 | 55.0% | Below GPT-6 Astra at 65.5% and Claude Opus 5.5 at 62.3% |
GraphWalks, up to 128K | 99.7% F1 | Highest of the four models in Google’s table |
GraphWalks, 256K to 1M | 84.2% F1 | Highest of the four models in Google’s table |
Chartography | 71.6% | Slightly above GPT-6 Astra at 71.0% |
CWE-bench v1 | 68.0% | Tied with GPT-6 Astra in Google’s table |
These are provider-reported launch results. Benchmark scores depend on model version, reasoning effort, sampling, tools, scaffolding and the exact dataset release. Use them to select candidates, then test Argon on frozen tasks from the intended workflow.
Pricing and access
Gemini 4 Argon is generally available through the paid Gemini API in Google AI Studio and Google Cloud under model ID gemini-4-argon. Google AI Pro and Ultra subscribers can access Argon in the Gemini app. Google AI Ultra subscribers can also use it in Google Antigravity. Enterprise users can access it through Gemini for business on web, desktop and mobile.
The launch price is temporary. Through November 30, 2026, API usage costs $2 per million input tokens and $10 per million output tokens. Starting December 1, the standard rates are $4 input and $20 output per million tokens. Cached input is priced at a 95% discount from the applicable input rate.
Usage or access item | Current value |
|---|---|
API model ID | gemini-4-argon |
Introductory input | $2 per million tokens through November 30, 2026 |
Introductory output | $10 per million tokens through November 30, 2026 |
Standard input | $4 per million tokens after November 30, 2026 |
Standard output | $20 per million tokens after November 30, 2026 |
Cached input | 95% discount from the applicable input rate |
Developer access | Paid Gemini API in Google AI Studio and Google Cloud |
Consumer access | Gemini app for Google AI Pro and Ultra subscribers |
Agentic coding | Google Antigravity for Google AI Ultra subscribers |
Fairwind access | Separate cyber-permissive model for approved defensive-security projects |
Token price is only one part of total cost. Include thinking tokens, cache behavior, tool calls, retries, wall time, failed-task recovery and human review. The useful comparison is cost per accepted result, not price per million tokens in isolation.
Limitations and deployment risks
- The one-million-token limit is shared across input, internal thinking and output, so every part of a long request competes for the same budget.
- The introductory token rates expire after November 30, 2026, when both input and output prices double.
- Google’s model card warns that Argon can hallucinate and may occasionally be slow or time out.
- The knowledge cutoff is March 2026, and knowledge may be older in some domains.
- The strongest launch benchmark results are provider reported and depend on Google’s selected harnesses, tools and settings.
- Long-running agents can drift from the original objective, follow injected instructions or take unacceptable intermediate actions.
- Cybersecurity capability is dual use. The standard model retains cyber safeguards, while the cyber-permissive version is restricted to approved defenders.
High-stakes legal, medical, financial, scientific and security work needs qualified review. Log the exact model version and configuration, separate untrusted content from system instructions, restrict tools to the minimum required scope and require approval before irreversible or externally visible actions.
How to evaluate Gemini 4 Argon
Build a frozen evaluation set of 20 to 50 real tasks. Include routine work, difficult edge cases, adversarial inputs, long-context examples and cases where the correct behavior is to stop or escalate. Compare Argon with Gemini 3.8 Flash and at least one neighboring frontier model using equivalent tools and source material.
Test area | What to record |
|---|---|
Task completion | Pass or fail against a written acceptance rubric |
Reliability | Repeated-run success rate, variance, timeouts and silent failures |
Factuality | Unsupported claims, source use, quotations and citation accuracy |
Tool use | Wrong calls, retries, recovery and permission-boundary failures |
Long context | Retrieval accuracy, instruction retention and total shared-token use |
Efficiency | Wall time, input, thinking, output, cached tokens, tool charges and review time |
Safety | Prompt injection, sensitive-data handling and irreversible-action controls |
Choose the least expensive configuration that clears the acceptance threshold with a safety margin. Re-run the suite when the model, system prompt, effort level, retrieval layer, tool definitions or approval policy changes.
Who should use it?
Situation | Recommendation |
|---|---|
Strong fit | Long-running software engineering, enterprise knowledge work, science and defensive cybersecurity |
Pilot first | Large contexts, computer use and workflows with several tools |
Escalate | Ambiguous or consequential work that does not reliably clear the evaluation threshold |
Avoid unsupervised use | Irreversible actions, sensitive data or high-stakes decisions without monitoring and approval |
A migration should be driven by measured outcomes. Keep Gemini 3.8 Flash available during the pilot, record where each model succeeds or fails, and use routing when different task classes have different quality and cost requirements.
Frequently asked questions
Is Gemini 4 Argon available now?
Yes. The standard Gemini 4 Argon model became generally available on October 13, 2026 through the paid Gemini API, Google AI Studio, Google Cloud, the Gemini app and selected Google products. Access depends on the product and subscription tier.
What is the Gemini 4 Argon API model ID?
Use gemini-4-argon. Google also documents a separate gemini-4-argon-cyber-permissive model for approved Fairwind projects.
How much does Gemini 4 Argon cost?
Through November 30, 2026, the introductory API price is $2 per million input tokens and $10 per million output tokens. After that date, the standard price is $4 input and $20 output per million tokens. Cached input receives a 95% discount.
How large is the Gemini 4 Argon context window?
Input, thinking and output share a unified window of up to one million tokens. A long prompt or extended reasoning trace therefore reduces the remaining output budget.
Is Gemini 4 Argon open source?
No. Gemini 4 Argon is proprietary. Access, serving behavior and lifecycle decisions are controlled by Google and supported distribution partners.
What is Gemini 4 Argon best used for?
Its strongest documented fit is long-running software engineering, enterprise knowledge work, science and defensive cybersecurity. Start with a supervised pilot and retain human sign-off for consequential work.
Should I migrate from Gemini 3.8 Flash?
Only after a side-by-side evaluation. Measure accepted-task quality, total cost, latency, output style, tool reliability and migration engineering. Argon is more capable on many provider evaluations, but it is also a frontier-priced model built for deeper work.
Related model guides
Browse the Best AI Models directory.
Compare with the Gemini 3.7 Flash guide.
Compare with the Gemini 3.1 Pro guide.
Compare with the GPT-6 Astra guide.
Official sources and update policy
Google’s Gemini 4 Argon general-availability announcement.
Google’s Gemini 4 Argon developer announcement.
Google DeepMind’s Gemini 4 Argon model card.
Google Cloud’s Gemini 4 Argon launch and pricing details.
Google Cloud’s Gemini 4 Argon developer guide.
Checked October 15, 2026. We update this guide when Google changes the specification, pricing, access, safety documentation or model lifecycle. Vendor benchmarks are attributed and are not presented as independent testing. No vendor payment or affiliate relationship determined inclusion.