Gemini 4 Argon: Pricing, Specs, Benchmarks and Who Should Use It

Gemini 4 Argon: Pricing, Specs, Benchmarks and Who Should Use It

AI Takeaway Teamon 2026-10-14

Last verified: October 14, 2026. All figures come from the official pages linked below; we have not run independent benchmarks.

Gemini 4 Argon is Google's new flagship reasoning model. On October 13, 2026, Google DeepMind announced that "Gemini 4 Argon is now globally available". Until then, access had been limited to trusted cyber defenders. The launch reached 1,042 points and 768 comments on Hacker News. This guide covers what the model is, what it costs (including a price rise on December 1), its specs and official benchmarks, how early testers reacted, and who should use it.

At a glanceGemini 4 Argon
DeveloperGoogle DeepMind
API model IDgemini-4-argon
StatusGlobally available (Oct 13, 2026); paid tier only in the Gemini API
Price per 1M tokens$2 input / $10 output through Nov 30, 2026; $4 / $20 from Dec 1, 2026
Context window1,048,576 tokens (unified input + output)
Max output65,536 default; up to 1,048,576 with long decode
InputsText, image, video, audio, PDF

Practical rule: if you are budgeting beyond November, use the December price ($4 / $20), not the launch price.

What Gemini 4 Argon is

Google's developer docs describe Argon as "our flagship reasoning model, built for complex software engineering, deep mathematical and scientific problem solving, autonomous multi-step agents, and long decode up to 1M tokens." The September 30 announcement positioned it for "real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense."

Google's model page lists internal uses, including:

  • Helping the Chrome team separate core webpage handling from a ~6 million line legacy folder
  • Making the xTB quantum chemistry package up to 2.3x faster at predicting molecular structures
  • Building a simulation from a raw video of cornhole, which taught Apptronik's Apollo humanoid to sink seven consecutive shots (DeepMind on X)

Benchmark chart for Gemini 4 Argon published by Google

Availability and pricing

Per the Gemini API pricing page, Argon "is not available on the Free tier or on the Google AI Pro plan." Paid-tier Standard prices per million tokens:

Through Nov 30, 2026From Dec 1, 2026
Input$2.00$4.00
Output (incl. thinking tokens)$10.00$20.00
Context caching$0.10$0.20
Batch / Flex input / output$1.00 / $5.00$2.00 / $10.00
Priority input / output$3.60 / $18.00$7.20 / $36.00

Caching is implicit only; explicit caching is not supported. Grounding with Google Search includes 5,000 free requests per month, shared across Gemini 3 and newer models, then costs $14 per 1,000.

Specs

From the model docs:

Inputs and limits

  • Text, image, video, audio and PDF in; text out
  • 1,048,576-token context window, shared by input and output
  • Up to 1,048,576 output tokens with long decode (65,536 by default)

Supported

  • Thinking at low, medium and high
  • Code execution, function calling, structured outputs and file search
  • Search and Maps grounding, and URL context
  • Computer use (Preview)
  • Batch, Flex and Priority inference

Not supported

  • Image generation, audio generation and the Live API

Official benchmarks

All figures below come from Google's model page (methodology), so they are vendor-reported:

BenchmarkGemini 4 ArgonGPT-6 AstraClaude Opus 5.5
Vals Index68.9%63.1%67.0%
AutomationBench51.3%41.4%42.5%
DeepSWE v1.177.9%74.1%74.2%
Terminal-bench 4.057.4%58.2%66.4%
Terminal-Bench Science 0.157.6%68.1%63.3%
GraphWalks 256k–1M (BFS F1)84.2%71.8%66.8%
LVBench (long video)91.7%87.5%83.7%
CWE-bench v168.0%68.0%67.0%

On Google's own table, Argon leads on knowledge work, long context and video. It trails on Terminal-bench 4.0 and on science tasks.

Vendor benchmarks show you what a model can do, not what it will do on your workload.

Early real-world reactions

Early reactions on X were mixed:

  • Ethan Mollick wrote that Gemini 4 "is the first non-OpenAI/Anthropic model to pass the Fable/Mythos/Astra bar" (post).
  • Aaron Levie said that on Box's eval it is "a huge upgrade over Gemini 3.1 Pro, jumping by 14 percentage points across all our tests" (post).
  • Bindu Reddy reported that on her team's new benchmark "Argon 4.0 falls short of GPT 6.1 Sol" (post).
  • Dan Shipper said Argon "comes in at 21%" on Every's Senior Engineer Benchmark (post).

Watch on YouTube: "I tested Gemini 4 Argon - does it beat Opus 5.5 and Astra?" by AI BROS

Third-party hands-on video by AI BROS, an independent creator. AI Takeaway did not make this video and is not affiliated with its creator.

No time to watch the whole review? AI Takeaway is a tool that works on any YouTube video, including third-party reviews like this one. Paste the link to get the key points in minutes, jump to the original timestamps, and ask follow-up questions; every answer cites the video. You can also browse the AI Takeaway video summary hub for more AI and tech interviews.

Who should use Gemini 4 Argon

  • Multimodal workflows: teams working with video, audio and PDFs. The model accepts all of them natively and scores 91.7% on LVBench per Google.
  • Huge outputs: anyone who needs very long outputs in one call, such as full-repo generation, long proofs or legacy migrations. Long decode goes up to ~1M tokens.
  • Security teams: Argon ties for first on CWE-bench v1, and Google is expanding its Fairwind Program for defenders.
  • Probably not for: cost-sensitive, high-volume simple tasks. The December price and the lack of a free tier make smaller models a better fit there.

How it compares with rivals (briefly)

Argon's launch price matches OpenAI's GPT-6.1 Sol: both list $2 input / $10 output, and both charge $0.10 for cached input. Sol takes only text and image input, with a 1.05M context and 128K max output. Neither vendor publishes a direct Argon-vs-Sol benchmark. Early independent tests above disagree on which is stronger, so test on your own tasks.

Getting started checklist

  1. Enable a paid Gemini API tier (Argon isn't on the Free tier).
  2. Call gemini-4-argon, starting at medium thinking.
  3. Estimate cost at both the current and the December 1 prices.
  4. Run 50–100 of your real tasks, and compare against your current model on cost per finished task.
  5. Re-check the pricing page before you commit.

Keep up without the watch time

Gemini 4 Argon and the next wave of models keep arriving with launch streams, interviews and reviews. If you would rather read than watch, the AI Takeaway video summary hub has free curated summaries of AI and tech interviews, for example:

To keep up with ongoing AI news, check the hub regularly. You can also install the AI Takeaway Chrome extension to get key points right on YouTube, or submit any YouTube video yourself on aitakeaway.net.

FAQ

Is Gemini 4 Argon free?

Not in the Gemini API. Google's pricing page says it is not available on the Free tier or on the Google AI Pro plan.

How much does Gemini 4 Argon cost?

$2 per million input tokens and $10 per million output tokens through November 30, 2026, rising to $4 / $20 on December 1, 2026.

What is Gemini 4 Argon's context window?

1,048,576 tokens, shared by input and output. Default max output is 65,536 tokens, and up to about 1M with long decode.

Is Gemini 4 Argon better than GPT-6.1 Sol?

Neither vendor publishes a direct comparison, and early independent tests disagree. On Google's own table, Argon scores 77.9% on DeepSWE v1.1 versus 74.1% for GPT-6 Astra.

Sources

Gemini 4 Argon: Pricing, Specs, Benchmarks and Who Should Use It