Is Gemini 4 Argon out?

Yes. Google announced Gemini 4 Argon on September 30, 2026 and broader global availability on October 13. Developer access includes the Gemini API and Google AI Studio; eligible Ultra and Pro subscribers can use the Gemini app and AI Mode. This page provides reAPI request examples and current token rates.

What model ID does Gemini 4 Argon use?

Use gemini-4-argon as the model identifier. Native requests put it in the model URL, while chat-compatible requests use the model field. Keep the identifier exactly as shown in the examples; Gemini 4 Argon with spaces is the display name.

Is Gemini 4 Argon free or paid?

Gemini 4 Argon uses a paid developer API tier. Google's October 13 announcement also provides Gemini app and AI Mode access for Ultra and Pro subscribers. Consumer subscriptions and API bills are separate; on reAPI, check the current token rates here and include thinking in your task estimate.

Does Gemini 4 Argon require a monthly subscription?

reAPI bills Gemini 4 Argon by token usage without a required monthly subscription. Consumer Gemini subscriptions are separate from API usage. Compare the current rates here with the input and generated tokens your workload actually uses.

How do I get a Gemini 4 Argon API key?

Create a reAPI API key in your dashboard and use it with the authentication shown in the Gemini 4 Argon documentation. Keep the key private in server-side configuration. Google-issued keys authenticate Google's own service; use the credential issued by the service you call.

What is the Gemini 4 Argon output limit?

One native JSON request returns up to 32,768 tokens and one streaming request up to 65,536. Gemini 4 Argon can continue beyond those caps within its shared 1,048,576-token window, which also contains input and thinking. Your client submits continuation requests; reAPI does not continue automatically on the server.

How does Gemini 4 Argon perform on benchmarks?

Google's September 30 table reports Gemini 4 Argon at 77.9% on DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on LVBench. These are dated, attributed results with evaluation conditions. The linked benchmark article explains the comparisons; measure your own tasks before choosing a model.

Is Gemini 4 Argon better than Claude Opus 5.5?

The task matters. Gemini 4 Argon leads Claude Opus 5.5 on DeepSWE v1.1 and several knowledge-work tests in Google's launch table; Claude leads on Terminal-bench 4.0 and PostTrainBench. Run the same prompts, tools and acceptance checks on both before treating a benchmark difference as a workflow advantage.

Can I run Gemini 4 Argon locally?

Google's cited developer pages describe Gemini 4 Argon as a hosted model and do not provide downloadable weights. Use a documented hosted endpoint for inference. A local application can call that endpoint, which is different from running the model itself on your computer.

Which Gemini 4 Argon thinking level should I use?

Start with medium, the native default. Gemini 4 Argon also supports low and high; compare them on the same task and choose the lowest setting that meets your requirements. Thinking cannot be fully disabled, and hiding thought summaries does not switch reasoning off.

Gemini 4 Argon — Google's Frontier Model for Complex Work

Gemini 4 Argon brings adjustable reasoning to coding, research and agent workflows. Give Gemini 4 Argon a clear task, work within its shared 1M-token context and choose the thinking level you need.

Highlights: Gemini 4 Argon: shared 1M-token context · adjustable thinking · text generation
Best for: Gemini 4 Argon for codebase migrations, document research and tool-based agents
Input: Gemini 4 Argon accepts text, images, video, audio and PDF, per Google
Output: Gemini 4 Argon generates text; 32,768 tokens per JSON request or 65,536 per stream

Model

gemini-4-argon

Fixed for this page. Playground requests cannot switch to another model.

Parameters

Leave a field blank to use its default.

generationConfig.maxOutputTokens

generationConfig.temperature

generationConfig.topP

generationConfig.seed

generationConfig.frequencyPenalty

generationConfig.presencePenalty

generationConfig.thinkingConfig.thinkingLevel

streamGenerateContent (SSE)

generationConfig.stopSequences

other fields

Parameters

other fields

A JSON object of any other request fields, sent as written.

Gemini 4 Argon· Gemini GenerateContent

Ask Gemini 4 Argon

Start with the prompt in the composer or choose an example.

Pricing

Credit-based — 1 credit = $0.001 USD. Pay only for completed generations.

First test cost

$0.00484.8 credits

One 1K-token prompt with a 1K-token answer at the cheapest tier.

Testing budget guide

Add credits

$10

≈ 2083 tests

$50

≈ 10416 tests

$100

≈ 20833 tests

Input tokens: $0.8; Price · 1M tokens
Output tokens: $4.0; Price · 1M tokens

What is Gemini 4 Argon?

Gemini 4 Argon is Google's reasoning model for complex coding, research and multi-step agent workflows. Announced on September 30, 2026 and released more broadly on October 13, it is documented in the Gemini API and Google AI Studio with the model ID gemini-4-argon. Gemini 4 Argon shares 1,048,576 tokens across input, thinking and output, so large source material leaves less room for generation. Longer work can continue across requests; reAPI returns each request separately and leaves continuation to your client. Use the Gemini 4 Argon examples and request reference to choose a format, then evaluate the same task at different thinking levels.

Gemini 4 Argon poster for codebase migration and research

What you can build with this model

Real-world workflows and production use cases you can build and ship with this model.

Gemini 4 Argon poster for long codebase migrations

Plan a migration with Gemini 4 Argon

Google's September 30 announcement describes Gemini 4 Argon agents migrating C and C++ code to Rust, including work on the Fuchsia Zircon kernel, with automated and manual review before shipping. Use Gemini 4 Argon to work through a module's dependencies, propose changes and identify the tests that should pass. For a large Gemini 4 Argon migration, keep source material within the shared context and handle continuation explicitly when generation needs another request.

Gemini 4 Argon poster for legal and finance research

Research documents with Gemini 4 Argon

Gemini 4 Argon scored highest on Vals Finance Agent v2 and Harvey's Legal Agent Benchmark in Google's September 30 comparison. Google documents image, video, audio and PDF input alongside text, extending Gemini 4 Argon beyond plain-text prompts. Give Gemini 4 Argon a specific research question and ask it to connect each conclusion to the supplied evidence; review the sources before relying on the result.

Gemini 4 Argon poster for defensive security reviews

Review security fixes with Gemini 4 Argon

Google introduced Gemini 4 Argon with examples of finding, validating and patching software vulnerabilities; it tied for first on CWE-bench v1 in Google's launch table. For an authorized code review, ask Gemini 4 Argon to explain the suspected flaw, propose a bounded fix and identify regression checks. The Gemini 4 Argon benchmark reflects Google's evaluation, while your own tests decide whether a proposed patch is ready.

Why reAPI for Gemini 4 Argon

Use your reAPI key with Gemini 4 Argon

Send Gemini 4 Argon requests with the same reAPI key you use for other supported models. Keep the key on your server and start with a small task whose answer you can check before expanding the workload.

Choose your Gemini 4 Argon request format

Gemini 4 Argon supports a native Gemini request surface on reAPI alongside chat-compatible requests. Use the native Gemini 4 Argon reference for thinking settings, structured output and function-call conversations; your application remains responsible for executing tools and sending their results back.

Budget Gemini 4 Argon by token usage

Gemini 4 Argon usage is billed by token, with current rates on this page. Include thinking and any continuation requests when comparing task costs; a short visible answer does not necessarily mean a small generation budget.

Gemini 4 Argon vs Claude Opus 5.5

Gemini 4 Argon leads on several coding, legal and video tasks in Google's September 30 comparison; Claude Opus 5.5 leads on terminal work. Compare the dated results below, then test the same workload and budget on both models.

Capability

Gemini 4 Argon

Claude Opus 5.5

Availability

Native Gemini and chat-compatible requests on reAPI

Available on reAPI now

Output and continuation

32,768 per native JSON request; 65,536 per stream; longer output via continuation

128K standard output limit in the referenced model specification

Context window

1,048,576 tokens shared by input, thinking and output

1M tokens

DeepSWE v1.1 (long-horizon coding)

77.9%

74.2%

Terminal-bench 4.0 (terminal work)

57.4%

66.4%

Harvey's Legal Agent Benchmark

19.6%

3.8%

LVBench (long video)

91.7%

83.7%

Scores are from Google's September 30, 2026 announcement; most non-Gemini scores were reported by their creators, with differing evaluation conditions. Gemini 4 Argon request limits follow Google's developer guides, checked October 14, 2026. Claude limits retain the specification used in the original comparison.

How to use Gemini 4 Argon

step 01
Create a key for Gemini 4 Argon
Create an API key in your dashboard and keep it in server-side configuration. Use that key to authenticate Gemini 4 Argon requests; a Google API key is a different credential.
Open
step 02
Try Gemini 4 Argon on a verifiable task
Start Gemini 4 Argon with a short prompt or a small source excerpt. Compare the Gemini 4 Argon answer against your acceptance criteria, then adjust the thinking level and generation budget for that workload.
Open
step 03
Connect Gemini 4 Argon to your application
Follow the Gemini 4 Argon request reference for native JSON, streaming and chat-compatible examples. Preserve function-call history and continuation information when your Gemini 4 Argon workflow needs multiple requests.
Open

docs/api/gemini-4-argon

API reference

Drop-in code and the full parameter table.

View full reference

curl https://reapi.ai/api/v1beta/models/gemini-4-argon:generateContent \
  -H "x-goog-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"role":"user","parts":[{"text":"Plan a C++ to Rust migration for this module"}]}],"generationConfig":{"maxOutputTokens":4096,"thinkingConfig":{"thinkingLevel":"medium"}}}'

Parameter validation failed

Rejected synchronously with the failing field named. Check enum values, ranges, and URL fields — nothing is charged.

Authentication or balance issue

401 means a missing or invalid Bearer key; 402 means the reserve exceeds your balance. Manage keys and credits in the console.

Content or material rejected

Reference material and outputs pass automated moderation. Rejected tasks fail with a clear error and a full refund.

Task failed or timed out

A task that reaches a failed state is never charged — the reserve refunds automatically. Keep the task id and retry when ready.

Frequently asked questions

Common questions about this model.

Gemini 4 Argon is Google's reasoning model for complex coding, research and multi-step agents. It produces text and has a unified 1,048,576-token window shared by input, thinking and output. Google documents multimodal input; the request reference explains the formats available through reAPI.