Gemini 4 Argon GA: Long Documents, Long-Running Ops, and a Price That Rises After Nov 30

Gemini 4 Argon GA: Long Documents, Long-Running Ops, and a Price That Rises After Nov 30

On October 13, 2026, Google made Gemini 4 Argon generally available. The pitch is frontier work on hard, multi-step jobs across coding, finance, legal, cybersecurity, and science, plus an industry-leading 1 million unified input-output token limit on the Gemini API.

For operators, two numbers matter more than the demo reel: the introductory API price, and the fact that price rises after November 30, 2026. If you are budgeting long proposals, contracts, or multi-file ops through November and December, write both prices down now.

What Google shipped

Google says Argon is globally available for:

• Developers and enterprises via the API in AI Studio and the Gemini platform
• Google AI Ultra subscribers in Google Antigravity
• Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search

Google Cloud's launch coverage lists the model id as gemini-4-argon. Google positions Argon for deep work on complex, real-world tasks and says it outperforms other frontier models on benchmarks across domains, including a claim of ranking #1 on the Vals Index. That ranking is Google's claim. Independent third-party confirmation beyond that claim was not verified for this article.

Intro price vs post-November 30 price

Google's launch post is explicit about the schedule:

• Introductory price: $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off
• After the introductory period expires on November 30, 2026: $4 per 1M input tokens and $20 per 1M output tokens

Google's footnote says the post-intro prices apply to the Argon model intended for public access. Do not build a December budget on the October intro rate.

That is the operator detail most launch posts bury. Argon is usable today at the lower rate. It is not a permanent $2 / $10 model unless Google extends the window.

Why the 1M token limit matters

Google states Argon has an industry-leading 1M unified input-output token limit via the Gemini API. For day-to-day ops, that is less about bragging rights and more about whether you can keep a long contract pack, proposal draft, and supporting files in one job without jumping tools midstream.

Long context is still not a substitute for good retrieval. It does change whether a finance or legal-style workflow can stay inside one model call while you review the output.

Vendor-reported impact numbers

Google published internal and early-access results. Treat them as Google's statements:

• Argon in Antigravity "helped teams increase agentic code submissions by more than 35% while reducing code rollbacks by nearly a third"
• On ADK Python work, Google says Argon "cut in-memory session prep time by 40% in long sessions and sped up cold imports by 21%"
• Embodied Reasoning (ERQA): Google states an 87% score and says Argon "outperforms all other frontier models" on pure reasoning for that benchmark
• Prompt injection: Google says Argon cut attack success across 15 adaptive attempts to 0.7% on Gray Swan IPI
• Agent shortcut attempts on coding benchmarks: "dropped to just 2% on Gemini 4 Argon"

Those figures are useful as signals of what Google optimized for: longer agent work, fewer unauthorized shortcuts, and stronger resistance to injected instructions. They are not your ROI study.

Google is also offering 10,000 scientists the highest usage tier in Google Antigravity free of charge for 6 months, subject to eligibility, capacity, and selection. That is a research access program, not a free production allotment for every business.

What this is not

Argon GA does not mean every long-running agent is ready to own money movement or customer sends without a gate. Google's own safety section emphasizes training the model to pause for human confirmation before irreversible steps and monitoring for shortcuts. That is a design goal, not a guarantee for your stack.

It also does not freeze pricing. If your finance model assumes $2 / $10 into Q1, the November 30 date breaks that assumption unless you renegotiate or switch models.

How to evaluate it this week

1. Pick one long-document or long-running job you already measure: contract review, multi-file proposal, or a multi-step ops agent.
2. Run it on Argon at the intro price and log tokens, wall time, and review edits.
3. Rebuild the same job's unit economics at $4 / $20 so December is not a surprise.
4. Keep a human gate on sends, payments, deletions, and anything that changes a system of record.

What it means for operators (Jaden's take)

I want a model that can hold a real business packet without falling over, and I want the price in writing. Argon is interesting because Google shipped both: a 1M unified token limit and a clear intro-versus-post-November-30 schedule.

My rule is simple. Use Argon where the work is long, checkable, and expensive for a person to assemble. Do not pretend the intro rate lasts forever. If a contract pack or multi-week ops agent gets better with Argon at $2 / $10, re-run the math at $4 / $20 before you wire it into December plans.

Budget both prices. Keep the human gate on irreversible actions. Trust your own edit log more than any #1 index claim.