Google Gemini 4 Argon: 1M Token Output at $2/10 USD

👁 120 reads

Google announced Gemini 4 Argon on September 30, 2026 at 10:02 PM Central European Time, its new flagship model. It offers one million tokens of output, beats GPT-6 Astra and Claude Opus 5.5 in most of Google’s tests, and will cost $2 per million input tokens and $10 per million output tokens at launch. For now, only cybersecurity specialists have access.

What Google Announced

The first post appeared at 10:02 PM on Sundar Pichai’s account, one minute later on Google DeepMind and Google accounts. The official statement from Google reads: “Today we’re introducing Gemini 4 Argon. It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit.”

Simultaneously, blog.google published a post by Koray Kavukcuoglu, SVP of Google DeepMind, and DeepMind released a full results table with methodology description. Key facts:

• Output limit: 1 million tokens in a single response, up from 64,000. GPT-6 Astra and Claude Opus 5.5 have 128,000 each. Google did not disclose the input window.
• Launch pricing: $2 / $10 USD per million input and output tokens, cache 95% cheaper. After the introductory period $4 / $20 USD, matching Claude Opus 5.5. GPT-6 Astra costs $10 / $50 USD.
• Access: Fairwind program only, meaning selected cybersecurity defenders, plus Google teams. Both groups receive a version without cybersecurity safeguards.
• Next in line: paid API customers and Google AI Ultra subscribers. No date given.

Benchmarks: 13 Wins Out of 19

Google compared Argon with GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across 19 table rows. We counted: Argon leads in 13, ties in one, Astra wins in three, Opus 5.5 in two. Selected results:

These are results provided by Google. Google calculated Argon’s scores itself, and took the others from public leaderboards and manufacturer data. Google did not show results for SWE-bench, GPQA, Humanity’s Last Exam, or ARC-AGI, and did not compare Argon with GPT-6.1 Sol. We collected the full table of all 19 tests in our Gemini 4 Argon guide.

How Google Is Already Using It

According to the post, thousands of Google employees use Argon daily. Agents built on this model freed over 300 TiB of storage in data centers, are rewriting code from C/C++ to Rust (up to 800,000 lines of the Zircon kernel), and sped up the libgav1 video decoder 2.7 times compared to the previous port. The company Wiz used it to find a critical vulnerability in hospital software that exposed personal data. These are manufacturer examples, not independent tests.

Reactions

Logan Kilpatrick from Google wrote on X primarily about pricing: “Argon is priced at $2 in and $10 out during introductory pricing!”. Varun Mohan, who builds Antigravity at Google, added: “Thousands of Googlers have already been using it in Antigravity internally”. Popular AI commentator @kimmonismus reacted: “HOLY, GEMINI 4 Released. And its on Fable / Astra level!!”, and @scaling01 commented on pricing: “only $10 output is also insane”.

There was also cold water. Bloomberg, cited by Techmeme, wrote that according to some Google employees, the model performs well in benchmarks but worse in some real programming tasks. Google disputes this assessment. The table only partially confirms this: in FrontierSWE v2 and Terminal-Bench 4.0, Argon clearly loses to the competition.

What We’re Doing With This

We have a ready set of tasks in our Lab where we tested GPT-6 Astra and Claude Opus 5.5. As soon as Argon reaches the API, we’ll run it on the same tasks and add the results. We’ll add the broader release date to our AI calendar. We’ve detailed the price and performance comparison with competitors in our article Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5, and the background of the entire family in our Gemini guide.

Frequently Asked Questions

When did Google announce Gemini 4 Argon?

September 30, 2026 around 10:02 PM Central European Time, with posts on X and blog.google.

How much does Gemini 4 Argon cost in the API?

Initially $2 USD per million input tokens and $10 USD per million output tokens, then $4 and $20 USD. Google did not specify how long the introductory pricing will last.

Who can use Gemini 4 Argon?

Currently only Fairwind program participants and Google teams. Next will be paid API customers and Google AI Ultra subscribers.

// read moreGoogle Gemini 4 Argon: New Model for Complex Professional TasksNewsGemini 4 Argon Pricing: API Rates, Hidden Token Traps, and CostsAI ToolsGemini 4 Argon vs GPT-6.1 Sol: Which Model to ChooseBenchmarks

Kinetyra