Gemini 4 Argon Pricing: What the December 1 Price Doubling Means for Your Budget
Google opened Gemini 4 Argon at $2 and $10 per million tokens, but only through November 30. Here is the math on what changes, and how to plan for it.
In this article 8 sections
Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026. On December 1, Google's standard price rises to $4 and $20, which doubles the bill for identical usage. Cached input gets a 95 percent discount. Teams should model both prices before they commit workloads.
Protocol Wire tracks what launches actually change for the people who ship software, and pricing windows are exactly the kind of detail that gets lost between a keynote and an invoice. This explainer is for product managers, engineering leads, and finance owners who budget for model usage. It includes a worked example you can adapt.
What Did Google Actually Announce?
Google made Gemini 4 Argon generally available on October 13, 2026, two weeks after announcing it on September 30 for a small group of vetted cyber defenders. According to Google's launch post, developers can reach it on the paid Gemini API tier, and enterprises can use it through Gemini for business.
Consumers are covered too. Google AI Ultra and Pro subscribers can use it in the Gemini app and AI Mode in Search. Individual access to Antigravity, Google's agentic coding tool, requires the Ultra tier. Free API accounts are excluded.
The model supports a 1 million token unified input-output limit through the API, with a continuation protocol for long outputs. Google's post does not describe a separate output cap, and VentureBeat noted that the company has not explained how that limit relates to the length of a single uninterrupted response.
How Much Will the Price Change Cost You?
The change doubles list prices, so any workload costs twice as much after November 30. The effect depends on your mix of input, cached input, and output tokens, so a worked example helps more than a headline.
Take a hypothetical monthly workload of 200 million input tokens and 40 million output tokens. Assume 120 million of the input tokens hit the cache. This table shows the bill at each price, assuming the cached discount stays at 95 percent after December 1.
| Token type | Volume | Through Nov 30 | From Dec 1 |
|---|---|---|---|
| Uncached input | 80 million | $160 | $320 |
| Cached input | 120 million | $12 | $24 |
| Output | 40 million | $400 | $800 |
| Total | $572 | $1,144 |
Two things stand out. Output tokens drive most of the bill, at about 70 percent, so verbose answers and long reasoning cost real money. And caching matters: the same 120 million cached tokens would cost $240 at the uncached launch price, so a stable prompt prefix saves far more than most teams expect.
How Does Argon Compare With Rival Models on Price?
At the introductory rate, Argon is cheaper than the premium tier of its competitors. VentureBeat's comparison lists Claude Sonnet 5.5 and GPT-6.1 Sol at $2 and $10, Claude Opus 5.5 and Meta's Muse Wonder at $4 and $20, and GPT-6 Astra at $10 and $50 per million tokens.
That puts Argon at half the cost of Opus 5.5 and one-fifth the cost of GPT-6 Astra until November 30. After December 1, it moves to the same list price as Opus 5.5 and Muse Wonder. At that point the question is quality per dollar, not the discount.
Quality is where the evidence splits. Vals AI ranks Argon first on its Vals Index at 68.90 percent. Artificial Analysis gives it an Intelligence Index of 53, tied with GPT-6 Astra and Claude Fable 5.1, and below Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Muse Wonder Full at 56. Google also reports that Argon trails Opus 5.5 on Terminal-Bench 4.0, 57.4 percent against 66.4 percent. Benchmarks like these are directional, so run your own tasks.
Why Do Output and Thinking Tokens Change the Real Cost?
Models differ in how many reasoning tokens they spend before answering, and those tokens bill at the output rate. Two models with the same list price can produce very different invoices on the same task. VentureBeat makes the same point, noting that reasoning-token use affects real cost.
That echoes our reporting on what AI infrastructure actually costs once the invoice arrives: the per-token price is only one line. Retries, long contexts, and tool calls all add volume. A cheaper model that needs two attempts can cost more than a pricier one that gets it right.
The practical fix is to measure cost per completed task, not cost per token. Log tokens in and out for a sample of real requests, record whether the result was accepted, and divide. That number survives price changes better than any list rate.
What Else Did Google Claim About Safety and Access?
Google says Argon led the Gray Swan Indirect Prompt Injection benchmark, cutting attack success to 0.7 percent across 15 adaptive attempts. Indirect prompt injection is when hidden instructions in a web page or document trick an agent into misbehaving. A low number is welcome, but it is a vendor-reported score on one test, so it does not replace your own red-teaming.
The rollout itself followed an unusual order. The model went first to vetted cyber defenders on September 30, and Google is now expanding access to hundreds more partners in its Fairwind Program. Qualified defenders can apply for access without cyber guardrails on a rolling basis. That staged approach gives security teams a head start, and it signals how vendors may handle dual-use models in future.
Google also announced that 10,000 scientists can apply for free access to Antigravity's highest usage tier for six months, subject to eligibility, capacity, and selection. That is a developer-relations move, not a pricing change, but it shows how the company is competing: low launch prices, free access for researchers, and distribution through its own apps and search.
For buyers, the takeaway is to read the access terms along with the price. A model that is cheap in the API but gated in the tool you actually use can still force a different budget than the headline suggests.
What Should Product Teams Do Before December 1?
Treat the date as a deadline for a decision, not a reason to burn tokens. A short plan covers most teams.
- Pull last month's usage and split it into uncached input, cached input, and output.
- Re-price the workload at the December rates using the table above.
- Run a side-by-side test of Argon, a cheaper model, and a stronger model on 50 real tasks.
- Move simple, high-volume work to the cheapest model that passes.
- Set spend alerts and per-project budgets in the API console.
Product decisions matter here, not only engineering ones. A feature priced per use can turn unprofitable overnight when a model bill doubles. If an AI feature already struggles with engagement, as described in our piece on AI features that go quiet after launch week, a higher cost per use makes the weak spots harder to justify.
Cost shifts also change what is worth building. When teams ship faster, as our piece on shipping velocity and prioritization explains, a model bill that doubles forces sharper choices about which features deserve inference spend.
Also avoid architecture that hard-wires a single vendor. Keep prompts and tool schemas portable, and route requests through a thin layer you control. Teams that built that layer can switch models in an afternoon when a price window closes.
Is Locking In Usage Before the Deadline a Good Idea?
Only for work you would do anyway. A temporary discount is a poor reason to run low-value jobs. Reprocessing an archive or generating bulk content just to use the cheaper rate can waste more than the saving.
The sensible exceptions are batch jobs with a real business purpose, such as backfilling document summaries or classifying a backlog you have been postponing. If the work is planned and the output will be used, doing it at $2 and $10 is a fair saving. Even then, validate quality on a small sample first.
Protocol Wire's editors read this launch as a pattern, not a one-off. Vendors will keep using time-limited prices to win trials, and buyers who track cost per task will make better choices than buyers who chase the cheapest headline.
One last habit helps: put the price change date in your team calendar, with a reminder two weeks ahead. Pricing windows are easy to forget until the first doubled invoice arrives, and a calendar entry costs nothing.
Conclusion: Budget for December, Not for October
The practical answer is simple. Model your workload at $4 and $20, because that is the price you will live with, and treat November's $2 and $10 as a bonus. Argon looks strong and competitively priced, but the evidence on whether it beats rivals is mixed.
Your next step is a one-week bake-off on real tasks, with cost per completed task as the scorecard. Do that before December 1 and you will know whether the new price is acceptable.
