![Gemini 4 Argon: 1M Tokens, $2 Price Doubles Dec. 1 [2026]](https://shattered.io/wp-content/uploads/2026/10/gemini-4-argon-malaysia-1m-tokens-price-doubles-dec-1-2026-gen.webp)
Google’s Gemini 4 Argon is now available worldwide through the Gemini API, and the pricing clock is already running. Google says the model carries a unified 1,048,576-token input-and-output context limit, costs $2 per 1 million input tokens and $10 per 1 million output tokens until November 30, 2026, and then doubles on December 1. Headlines in Southeast Asia, including coverage framed around ProductNation Malaysia, have pitched the launch as Gemini 4 Argon reaching Malaysia with paid plans only. That framing needs unpacking, because the official material supports parts of it and leaves other parts open.
This analysis, dated October 24, 2026, separates what Google has confirmed from what readers have inferred. We cover the price schedule, the token limits, the thinking controls, who can actually use the model, and what the December step-up means for budgets. For a head-to-head against OpenAI’s flagship, see our earlier Gemini vs ChatGPT comparison. This piece focuses on the Malaysian angle and the money.
Don't miss new tech stories on Google
Add Shattered once in the Google app and our stories show up in your news suggestions.
What Google Confirmed About Gemini 4 Argon #
Google’s official model material gives the product name as Gemini 4 Argon and the API identifier as gemini-4-argon. Developers who want to test it call that string, nothing more exotic. Google describes the model as globally available through the Gemini API for developers and enterprises. It also says Google AI Ultra and Pro subscribers can reach it in specified Gemini experiences, meaning the consumer-facing app is not the only door.
The headline spec is the context window. Google lists a unified limit of 1,048,576 tokens for input and output combined. That figure is exactly 2 to the 20th power, the usual 1 million shorthand in engineering terms. The default synchronous API output cap is far lower at 65,536 tokens, though Google says the model can produce up to 1,048,576 tokens with long decode.
Google also exposes three tunable thinking levels: low, medium and high. Thinking tokens are billed as output tokens, which matters for anyone watching the invoice. Google has not attached a speaker-attributed quotation to the launch in the excerpts we reviewed, so this article paraphrases the company’s published documentation and does not put words in any executive’s mouth.
The Malaysia Question: What the Evidence Supports #
The tagline that Gemini 4 Argon reached Malaysia is a regional framing, not a Google statement. Google’s own material says the model is globally available and lists an Asia-Pacific Google Cloud region among its serving locations. It does not name Malaysia, and it does not spell out any Malaysia-specific consumer rollout, local pricing or local-currency billing.
So the careful reading runs like this. A developer in Kuala Lumpur or Penang can almost certainly call the Gemini API under the global availability language, subject to the usual account, billing and regional terms. A Malaysian consumer hoping to find Argon inside the Gemini app on a free account faces a different picture, because Google ties consumer access to the AI Ultra and Pro plans.
We flag the gap because search traffic around this story leans on the Malaysia wording. Readers want to know whether they can sign up today, what it costs in ringgit and whether a local tier exists. On the evidence available, Google has answered only the first half of the first question. Anything about ringgit pricing, local telco bundles or Malaysian data residency is unconfirmed, and we won’t guess at it.
Gemini 4 Argon Pricing: $2 and $10 Through November 30 #
Google’s introductory API pricing is $2 per 1 million input tokens and $10 per 1 million output tokens, with output counting thinking tokens. Cached input gets a 95% discount, landing at $0.10 per 1 million tokens. The introductory period ends November 30, 2026. From December 1, 2026, Google’s schedule moves to $4 per 1 million input tokens, $0.20 per 1 million cached input tokens and $20 per 1 million output tokens.
| Line item | Through Nov. 30, 2026 | From Dec. 1, 2026 | Change |
|---|---|---|---|
| Input, per 1M tokens | $2 | $4 | 2x |
| Cached input, per 1M tokens | $0.10 | $0.20 | 2x |
| Output (incl. thinking), per 1M tokens | $10 | $20 | 2x |
| Cached-input discount vs. input | 95% | 95% | Unchanged |
| Free tier access | Not available | Not available | Unchanged |
Source for all figures above: Google’s Gemini API model documentation and launch material. Two details stand out. First, every price line doubles, so the ratios stay put. Output costs five times input, and cached input costs 5% of fresh input, before and after. Second, the cached discount holds at 95% across the change, which means caching strategies built now keep their relative value later.
The December 1 Price Step: What Doubling Costs You #
A uniform 2x increase is easy to state and easy to underestimate. Teams that prototype in October and November at $2 and $10 will see their production bill double the day they flip to steady-state traffic in December. The table below runs illustrative workloads through Google’s published rates. The workloads are our own arithmetic examples and do not describe any real customer.
| Illustrative workload | Cost through Nov. 30 | Cost from Dec. 1 |
|---|---|---|
| Support bot: 5M input, 1M output | $20.00 | $40.00 |
| Document review: 50M input, 5M output | $150.00 | $300.00 |
| Cached retrieval: 100M cached, 10M fresh input, 5M output | $80.00 | $160.00 |
| One full-window call: 983,040 input, 65,536 output | $2.62 | $5.24 |
| Long decode: 48,576 input, 1,000,000 output | $10.10 | $20.19 |
The cached retrieval row shows why caching matters. At $0.10 per 1 million tokens, 100 million cached tokens cost $10 in the introductory window, against $200 if the same tokens were billed as fresh input. After December 1 the cached cost rises to $20. Output can still dominate a bill when answers run long, since each output token costs five times an input token.
Because thinking tokens count as output, a high thinking level can inflate cost quickly. A short answer that triggers long internal reasoning bills for the reasoning. Budget owners should measure real token counts per request at each thinking level before committing to a December forecast.
The 1,048,576-Token Context Window in Practice #
A million-token window changes which jobs fit in a single request. Whole code repositories, long contract bundles, multi-hour transcripts and large research collections can sit in one prompt instead of being chopped into retrieval chunks. Google’s unified limit means input and output share the same 1,048,576-token budget, so a prompt that fills the window leaves no room for a reply.
What a full window costs
Filling the entire window with input costs about $2.10 at the introductory rate and about $4.19 after December 1. That is cheap for a single call and expensive at scale. A pipeline that sends 10,000 near-full-window requests a day would spend roughly $21,000 daily on input alone at $2, and roughly $42,000 at $4. Cached input trims those numbers sharply when the same large context gets reused across calls.
Google is not alone in chasing long context. Our earlier look at Kimi K3.1 Pro’s 1M context against Claude Opus 5.5’s pricing shows the same race playing out among rivals. Window size has stopped being a differentiator by itself. Price per useful token, recall quality deep into the window and latency now decide purchases, and Google has not published recall figures for Argon in the material we reviewed.
Output Limits: 65,536 by Default, Up to 1,048,576 With Long Decode #
Google sets the default synchronous API maximum output at 65,536 tokens, which equals 6.25% of the full 1,048,576-token ceiling. Developers who need longer generations must use the long-decode option, which Google says can reach the full limit. The synchronous cap protects latency and cost for ordinary calls, while long decode serves jobs such as book-length drafts, large code migrations and exhaustive structured extraction.
The practical advice is simple. Set an explicit output limit on every call. A runaway generation near the long-decode ceiling would cost about $10 at the introductory output rate and about $20 after December 1, per request. Multiply that by retries and parallel agents and a single bug can burn through a monthly budget overnight.
Long decode also raises reliability questions that Google’s excerpts do not answer. Timeouts, streaming behavior and partial-output handling all matter once responses run to hundreds of thousands of tokens. Teams should test those paths before they depend on them.
Thinking Levels: Low, Medium and High #
Gemini 4 Argon lets developers pick a thinking level of low, medium or high. Google positions this as a control over how much reasoning the model does before it answers. Higher settings generally trade speed and cost for depth, and because thinking tokens bill at the output rate, the setting is also a pricing lever.
A sensible rollout plan starts at low for classification, routing and simple extraction. It reserves medium for general drafting and analysis, and saves high for multi-step problems where a wrong answer costs more than the extra tokens. Teams should log reasoning token counts per level for a week and compare them to answer quality on their own test set. Google has not published a token-per-level table in the excerpts we saw, so every shop has to measure its own.
For coding workloads, the same discipline applies to the tools wrapped around the API. Our guide to Cursor models pricing walks through how per-model costs show up in a developer’s monthly bill, and the lesson transfers directly to a new model with a price cliff.
Free Tier Excluded, but Paid Plans Only Overstates It #
Google states that Gemini 4 Argon is not available on the Free tier of the Gemini API. That part of the viral framing is accurate. The paid plans only shorthand goes too far, though, because Google’s material also describes a limited free-of-charge six-month opportunity for eligible scientists and researchers who apply. Free access exists for a narrow group, just not as an open tier.
| Claim | Status | Basis |
|---|---|---|
| Globally available through the Gemini API | Confirmed | Google launch material |
| Available to Google AI Ultra and Pro subscribers in specified experiences | Confirmed | Google launch material |
| 1,048,576-token unified context limit | Confirmed | Google model documentation |
| No Free-tier API access | Confirmed | Google model documentation |
| Limited six-month free opportunity for eligible scientists and researchers | Confirmed, by application | Google program material |
| Malaysia-specific launch or pricing | Unconfirmed | Not named in official excerpts |
| Paid plans only as a universal rule | Unconfirmed | Researcher program contradicts it |
The table shows the pattern. Hard specs and pricing come straight from Google. The regional and universal-paywall claims come from headline shorthand. Readers who plan budgets around the first group will be fine, and readers who plan around the second may be surprised.
Who Gets Access: API, AI Ultra and AI Pro #
Three paths lead to Argon. Developers and enterprises use the Gemini API with a billing account. Consumers and prosumers use a Google AI Ultra or Pro subscription, which unlocks the model inside specified Gemini experiences. Eligible scientists and researchers can apply for the limited free six-month window.
The split matters for Malaysian readers. A student or hobbyist on a free Google account does not get Argon through the API Free tier. A freelancer building a client tool can pay as they go. A university lab could apply under the researcher route. Google has not published, in the material we reviewed, the monthly subscription price of AI Pro or AI Ultra for Malaysia, so we do not quote one.
Enterprise buyers face a different calculation. They care about regional processing, contractual terms and committed-use discounts, none of which appear in the launch excerpts. Procurement teams in Malaysia should ask Google Cloud directly about data location and billing currency before signing anything.
Market Impact for Malaysian and Southeast Asian Developers #
For developers across Southeast Asia, the combination of a million-token window and $2/$10 introductory pricing lowers the cost of building document-heavy products. Think multilingual contract review, customer-service archives and government-form processing. Long context reduces the engineering needed for retrieval pipelines, which suits small teams without dedicated infrastructure staff.
The December doubling pushes in the opposite direction. Startups pricing a product in ringgit while paying Google in dollars carry currency risk on top of the 2x step. A team that sets subscription prices in November on $2/$10 economics could see gross margin squeezed a month later. That timing risk is the real story for small builders, more than any single spec.
Supply is part of the backdrop. Sundar Pichai has said Google’s compute remains tight into 2027, a point we covered in our report on Pichai’s compute shortage remarks. Our reading, offered as analysis and not as a Google statement, is that an introductory price followed by a step-up fits a capacity-constrained market. Google has not said why it chose these dates or numbers.
Competitive Comparison: Where Argon Sits Against Rivals #
Our earlier head-to-head found the introductory Argon price lines up with the same $2/$10 headline pricing on OpenAI’s side, which makes the December change the interesting moment. If Google moves to $4/$20 while rivals hold, Argon becomes the premium option. If rivals follow, the whole tier reprices upward together. We cannot say which, and nothing in Google’s launch material hints at it.
On the Anthropic and OpenAI front, our Claude vs GPT agents comparison covers how those labs pitch their flagship models for agent work. Argon’s differentiators on paper are the unified 1,048,576-token window, the three thinking levels and the 95% cached discount. Whether those beat a rival’s agent tooling depends on workload, and we have no independent Argon benchmarks to cite today.
Open-weight models offer a third path for cost-sensitive teams. Our roundup of the best open-source LLMs in October 2026 lists options that start from $0.15 input pricing, which sits far below Argon’s $2. The trade is capability and context against price, and each team has to run its own tests to see where the line falls.
Historical Context: Google’s Gemini Cadence #
Google has shipped Gemini releases at a quick pace through 2026. We covered Gemini 3.6 Flash arriving in Antigravity with 17% fewer tokens earlier this season, a release aimed at efficiency rather than raw scale. Argon pushes the other direction with a larger window and a higher price tier. Together the two releases show Google working both ends of the market: cheap and fast on one side, long-context and reasoning-heavy on the other.
Introductory pricing that later steps up is not new in the industry. Providers use it to pull developers in, collect usage data and size demand before setting a long-run rate. What stands out here is the clean 2x across every line and the explicit end date. Google published both, so buyers can plan around them, which is better than a surprise repricing.
For broader reading on where models, benchmarks and platform shifts fit together, our AI and machine learning section tracks the releases that matter week to week.
Five Predictions for Gemini 4 Argon #
These are our own forecasts, based on the published schedule, and not statements from Google or any named executive.
- November usage spikes. Teams will pack testing and bulk jobs into the cheap window before November 30, since batch work done at $2/$10 costs half of what it will cost on December 1.
- Caching becomes standard. With a 95% discount that survives the price change, developers who reuse large contexts will adopt context caching as a default pattern.
- Medium becomes the default thinking level. Low will serve routing and extraction, high will serve a minority of hard tasks, and cost dashboards will track the mix.
- Regional clarity follows the buzz. The gap between the reached Malaysia headlines and Google’s global wording will push Google or its partners to publish clearer regional guidance, though we cannot predict when.
- Rivals respond on price or context. If Argon holds at $4/$20 after December 1, expect competitors to market against it on cost, and expect open-weight alternatives to gain interest from price-sensitive teams.
What Readers Should Do Before November 30 #
Start by measuring. Run a representative slice of your traffic through gemini-4-argon at each thinking level and record input, output and reasoning token counts. Price the result at both rate cards. If the December figure breaks your margin, you have a month to redesign prompts, add caching or route easy requests to a cheaper model.
Monthly cost = (fresh_input_M x input_rate)
+ (cached_input_M x cached_rate)
+ (output_M x output_rate)
Through Nov. 30: input $2, cached $0.10, output $10
From Dec. 1: input $4, cached $0.20, output $20
Next, set hard output caps and spend alerts. Then check eligibility paths: researchers should look into the six-month free program, and consumers should compare the AI Pro and AI Ultra plans against pay-as-you-go API costs for their actual usage. Finally, if you operate in Malaysia and need local billing or data-residency terms, ask Google Cloud for them in writing. The launch excerpts do not provide those answers.
A budgeting checklist for the price step
Finance and engineering leads can turn the schedule into a short routine. Pull last month’s token counts by request type. Reprice them at $4 input and $20 output. Compare the result against revenue per request. Where the margin turns negative, decide now whether to raise prices, cut context length, enable caching or move that traffic to a cheaper model.
Two more habits pay off. Tag every API call with a feature name so you can see which product line drives the bill. And alert on daily spend at 50% and 80% of budget, because a million-token context makes it easy to overspend by accident. A single misconfigured loop that sends full-window prompts costs about $2.10 per call at the introductory input rate, which adds up fast at machine speed.
How We Reviewed the Launch Material #
Every specification and price in this article comes from Google’s own launch and model documentation as reviewed on October 24, 2026. We link only to pages on this site and cite no independent benchmarks, because none were part of the verified material. Where Malaysian coverage went beyond what Google states, we labeled the claim unconfirmed and left it out of the facts. We will update this analysis if Google publishes regional availability details, local pricing or recall measurements for the full 1,048,576-token window.
FAQ: Gemini 4 Argon in Malaysia, Pricing and Access #
Is Gemini 4 Argon available in Malaysia?
Google describes the model as globally available through the Gemini API, and its material lists an Asia-Pacific Google Cloud region. It does not name Malaysia or announce a Malaysia-specific launch, so treat the Malaysia framing as unconfirmed shorthand.
How much does Gemini 4 Argon cost?
Through November 30, 2026, it costs $2 per 1 million input tokens and $10 per 1 million output tokens, including thinking tokens. Cached input costs $0.10 per 1 million tokens. From December 1, 2026, prices rise to $4, $0.20 and $20 respectively.
Is there a free tier for Gemini 4 Argon?
No. Google says the model is not available on the Free tier. A limited free-of-charge six-month opportunity exists for eligible scientists and researchers who apply.
What is the context window?
The unified input-and-output limit is 1,048,576 tokens. The default synchronous API output maximum is 65,536 tokens, and Google says long decode can reach up to 1,048,576 output tokens.
What are the thinking levels?
Developers can choose low, medium or high. Thinking tokens count toward output pricing, so higher levels can raise the bill.
Which subscriptions include Argon?
Google says Google AI Ultra and Pro subscribers can use the model in specified Gemini experiences. Google has not listed Malaysian subscription prices in the material we reviewed.
What is the API model identifier?
The identifier is gemini-4-argon.
Does the discount on cached input change in December?
The percentage does not change. Cached input stays 95% below fresh input, which means $0.10 now and $0.20 from December 1, because the base input price doubles.