Field SOP
Field SOP

$2 to $4 After November 30: Will Your Bill Double?

Gemini 4 Carbon's introductory price is $2 per million input tokens and $10 per million output tokens, same as Argon, and the vendor footnote states verbatim that after the introductory period expires on November 30, 2026 the price becomes $4 and $20 per million, applying to the Carbon model intended for public access only - the widely echoed December 1 date is a paraphrase; the official text gives only the expiry. This piece is a cost-accounting and decision checklist, not a setup tutorial: what the vendor actually said covers building today on the paid tier in Google AI Studio and Android Studio, consulting the Gemini API Developer Guide, and the introductory window through November 30; an explicit unstated list records what the pages do not specify (non-paid tiers, consumer timing, rate limits, context and tiered pricing, the verbatim model ID), with unstated never rendered as disallowed. The core claim follows the vendor cost qualifier - Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon uses more tokens - so a low unit price per token is not a low cost per task, since different models consume different token counts for the same job; benchmark ranks and cost-per-task comparisons are Google-reported and not independently audited. Angled against the October 30 Carbon hotspot, staying on price-and-cost wording only.

Published November 12, 20269 min read
<!-- gemini-4-carbon-intro-price-window-sop | sop | $2 to $4 After November 30: Will Your Bill Double? -->

A price tag, doubled: per million input tokens from $2 to $4, per million output tokens from $10 to $20, the introductory period expiring on November 30, 2026. The reflex is to multiply your usage by two and conclude your costs will double. But is this bill one multiplication? This piece is a cost-caliber checklist, not an integration tutorial: no button for switching models, no verdict on whether to switch. It sets out what the vendor stated, what it left unstated, and the boundary that a low unit price per token does not mean a low cost per task, so anyone on the API before November 30, 2026 can audit the numbers. Figures and wording were checked against the vendor release page and developer announcement dated November 10, 2026; the time anchor is as of November 11, 2026, with sources woven into the prose rather than listed at the end.

The site's Gemini 4 Carbon hotspot from October 30 covered a different gap: a top score with no way to buy. This piece stays with the price-and-cost caliber boundary in the November 10, 2026 pages. In under two weeks the model moved from unbuyable to paid-tier availability, with the introductory expiry written into a footnote; the easiest misreading is to treat a doubled unit price as a doubled bill. The price tag is one ledger; the bill carries a second, usage-dependent one.

What the Vendor Actually Said: The Footnote, Verbatim

The pricing footnote on the vendor release page reads, verbatim: After the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply. Applies to the Carbon model intended for public access only. In plain terms, after expiry the rate becomes $4 per million input tokens and $20 per million output tokens, and that rate Applies to the Carbon model intended for public access only. The public access only qualifier must be read on the same line as the price, never dropped.

Some write-ups paraphrase this into a dated start for the new rate, but the verbatim text only says expires on November 30, 2026: the introductory period ends that day, with no separate start date. This piece keeps the expiry wording and does not copy the paraphrase. The word introductory is just as easy to lose: the footnote says introductory period and the launch rate says introductory price, so $2 / $10 is a promotional tier, not a standing rate.

The launch rate, verbatim, is at the same introductory price as Gemini 4 Argon ($2 per million input tokens and $10 per million output tokens): during the introductory period, Gemini 4 Carbon costs $2 per million input tokens and $10 per million output tokens, the same as Argon. Before November 30, 2026 the two are identical on unit price; what separates them is token consumption, not this table. For a side-by-side of unit-price fields, see the site's LLM API pricing comparison; this piece fixes the caliber of this one price line.

The availability wording, verbatim, is available via the API in Google AI Studio and Gemini for business for all paying developers and enterprises, and It is also rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search. The API is open to paying developers and enterprises through Google AI Studio and Gemini for business; Google AI Ultra subscribers see it roll out across Google Antigravity, the Gemini app, and AI Mode in Search. Scientists get free of charge for 6 months, onboarded on a rolling basis, subject to eligibility, capacity, and selection; cybersecurity goes through Apply to join the Fairwind Program, also rolling. Two cautions: the page says available and rolling out, not GA or general availability, so this is a staged push; the October 28 post carried no expiry date, with November 30, 2026 appearing only in the November 10, 2026 footnote, so do not fold the two into one source.

The developer announcement keeps the stated actions to three: start building with the paid tier of the API in Google AI Studio and Android Studio; Google AI Ultra subscribers can try it in AI Studio and Google Antigravity; and Check out the Gemini API Developer Guide for model details and setup instructions. That the introductory rate runs until November 30, 2026 is stated plainly too. The vendor published no checklist of actions to take before that date, a point that matters below: the stated part ends here.

Why a Low Unit Price per Token Is Not a Low Cost per Task

The vendor wrote this boundary into the body text: Argon remains the ideal choice for the majority of software engineering and knowledge work; Carbon is built for complex, multi-step coding and debugging and open-ended scientific research and discovery, using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops. Read more tokens alongside the price sheet: the same unit price per token, multiplied by a larger token count, does not necessarily yield a lower cost per task.

The line this piece most wants to establish: a low unit price per token does not mean a low cost per task. A price sheet gives a unit price; a bill is that price multiplied by usage. Carbon and Argon do not consume the same tokens for the same task, and the vendor says Carbon uses more tokens. With both factors moving, inferring total cost from one does not hold. Nor can you compare Carbon's $2 / $10 with another model's unit price and conclude the total is cheaper: different models consume different token counts for the same task, so a cross-model unit-price comparison cannot establish total cost.

The vendor also offers numbers that appear to support "cheaper," but their dimension and provenance matter. The self-reported line, verbatim, is It ranks #1 on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench at more than 2X lower cost per task compared to the next best external model, and the MKA benchmark gives Carbon ranks #2 while being more than 6X lower in average cost per task compared to the top performing model. What is measured is cost per task, not price per million tokens. Both figures are Google-reported and not independently audited; the benchmark names come from third parties such as Cursor, Perplexity, and Zapier, but the word independent in the vendor's sentence is vendor wording, not proof of an audit. Rankings and cost per task comparisons are self-reported caliber, not objective conclusions, and should not be padded with absolute superlatives. The vendor leaves its own gaps: prompt injection robustness remains an area for continued improvement, broader benchmarking is still underway, and Details will be shared after the responsible disclosure process completes, so the self-reported numbers are not a sealed verdict.

What the Vendor Did Not Specify: Separating "Not Written" From "Not Allowed"

Set the discipline first: not specified by the vendor is not the vendor announcing absence, and certainly not a closed door. Every item below has no verbatim statement on the November 10, 2026 pages; this piece writes "not specified by the vendor," supplies no answer on the vendor's behalf, and never rewrites it into "not available."

The not-specified list: whether non-paying developers and enterprises can use it; when non-Ultra consumers get access; the completion time of rolling out; the price of Carbon for cyber; a formal GA date for /swarm; quotas and rate limits; context window length; maximum output tokens; tiered and cached pricing; the verbatim model ID spelling; free-tier availability; and whether the rate changes again after November 30, 2026. The developer announcement does not mention context, maximum output, modality, thinking, quotas, the free tier, or a price-change date, and independent pricing-page caliber was not verified, so these fields stay blank: no speculative figures, no outside paraphrase of context window, maximum output, tiered, or cached pricing presented as vendor wording.

The model ID deserves its own note. The query string of an AI Studio link shows gemini-4-carbon, but that is a link parameter, not a vendor-declared model ID. None of the three vendor pages spells out a model ID verbatim; the guidance touching model details is to check the Gemini API Developer Guide. Copying a link parameter into config as an official ID is the common integration mistake; this piece supplies no ID spelling and invents no buttons, menus, console paths, or screenshot-level entry points, leaving the final word to the official Developer Guide. The vendor pages neither say the free tier is closed nor open; neither assertion is vendor text.

One more category belongs outside the vendor column. Press coverage suggests running the same task on Argon and Carbon before November 30 to compare, stockpiling usage early, switching models, or changing billing tiers. Those are media suggestions, not vendor guidance; the vendor's stated actions are the three above. This piece does not recast media moves as vendor instructions, and does not talk you out of your own comparison: that is your decision, just not one to file under vendor advice.

What You Can Check Before November 30, 2026

This is a cost-caliber checklist, so what follows lists verification dimensions only: no buttons, no menus, no screenshot-level paths, no conclusion handed to you. The dimensions compress into a few lines; fill the rest with your own usage:

text
Price line: $2 per million input / $10 per million output during the introductory period; $4 / $20 afterward, public access only read on the same line
Usage line: input tokens, output tokens, and verification loops per task
Math line: unit price per token times token consumption equals cost per task; do not judge by unit price alone
Date line: November 30, 2026 is the introductory expiry; later repricing is not specified by the vendor
Check line: model ID and setup instructions come from the Gemini API Developer Guide, not a link parameter

The price line is the one the vendor fixes outright: $2 per million input and $10 per million output during the introductory period, the same as Argon; $4 / $20 after November 30, 2026, with public access only on the same line. Copy it into your cost model as-is. The usage line is most worth measuring yourself: how many input tokens, output tokens, and verification loops one task triggers on Carbon. The cost qualifier says Carbon uses more tokens but gives no multiple, so only your call logs can answer; the site's LLM API cost optimization SOP works from this usage side. The math line joins the first two: read cost per task as unit price per token times actual token consumption, not as a verdict from unit price alone. The date line reminds you that November 30, 2026 is the vendor-stated expiry and later repricing is not specified, so leave a budget margin. Self-reported cost per task multiples belong here only as a qualified hypothesis to verify, not a budget coefficient.

Fill in those lines and you get a ledger you can audit, not a switch verdict. The price tag comes from the vendor, the usage from you, and the bill is the product of both: a low unit price per token does not mean a low cost per task.

FAQ

Q1: Does the rate jump to $4 / $20 on December 1?

A1: The verbatim vendor text only says the introductory period expires on November 30, 2026; it does not name December 1, a media paraphrase this piece does not copy. After expiry the rate is $4 per million input tokens and $20 per million output tokens, with Applies to the Carbon model intended for public access only, so the public access only qualifier stays on the same line as the price.

Q2: Can the free tier use Gemini 4 Carbon?

A2: The vendor-stated scope is the paid tier for paying developers and enterprises plus Google AI Ultra subscribers; free-tier availability is not specified by the vendor. That is not the vendor announcing a closed door, so this piece draws no conclusion; availability is whatever you see in Google AI Studio and the Gemini API Developer Guide.

Q3: What is the vendor-declared model ID?

A3: None of the three vendor pages spells out a model ID verbatim. The gemini-4-carbon in an AI Studio link query is a link parameter, not a vendor-declared model ID; the related guidance is Check out the Gemini API Developer Guide for model details and setup instructions. This piece offers no ID spelling and invents no buttons or menu paths.

Q4: Should I run a comparison, stockpile usage, or switch models before November 30?

A4: Those are media suggestions, not vendor guidance. The vendor's stated actions are build with the paid tier, check the Gemini API Developer Guide, and the introductory rate until November 30, 2026. This is a cost-caliber checklist giving verification dimensions only; it decides no switch and recasts no media move as a vendor instruction.

Q5: Can the self-reported rankings and cost per task figures be treated as conclusions?

A5: No. The three rankings, the more than 2X lower cost per task, and the MKA ranking and multiple are Google-reported and not independently audited, measuring cost per task rather than unit price. A low unit price per token does not mean a low cost per task, the vendor says Carbon uses more tokens, and a cross-model unit-price comparison cannot establish total cost. Treat them as a hypothesis to verify yourself.

Related reading

This article is AI-assisted and human-edited. Last updated: 2026-11-12

FAQ

Does the rate jump to $4 / $20 on December 1?
The verbatim vendor text only says the introductory period expires on November 30, 2026; it does not name December 1, a media paraphrase this piece does not copy. After expiry the rate is $4 per million input tokens and $20 per million output tokens, with Applies to the Carbon model intended for public access only, so the public access only qualifier stays on the same line as the price.
Can the free tier use Gemini 4 Carbon?
The vendor-stated scope is the paid tier for paying developers and enterprises plus Google AI Ultra subscribers; free-tier availability is not specified by the vendor. That is not the vendor announcing a closed door, so this piece draws no conclusion; availability is whatever you see in Google AI Studio and the Gemini API Developer Guide.
What is the vendor-declared model ID?
None of the three vendor pages spells out a model ID verbatim. The gemini-4-carbon in an AI Studio link query is a link parameter, not a vendor-declared model ID; the related guidance is Check out the Gemini API Developer Guide for model details and setup instructions. This piece offers no ID spelling and invents no buttons or menu paths.
Should I run a comparison, stockpile usage, or switch models before November 30?
Those are media suggestions, not vendor guidance. The vendor's stated actions are build with the paid tier, check the Gemini API Developer Guide, and the introductory rate until November 30, 2026. This is a cost-caliber checklist giving verification dimensions only; it decides no switch and recasts no media move as a vendor instruction.
Can the self-reported rankings and cost per task figures be treated as conclusions?
No. The three rankings, the more than 2X lower cost per task, and the MKA ranking and multiple are Google-reported and not independently audited, measuring cost per task rather than unit price. A low unit price per token does not mean a low cost per task, the vendor says Carbon uses more tokens, and a cross-model unit-price comparison cannot establish total cost. Treat them as a hypothesis to verify yourself.

Related

Field SOP

GitHub Copilot: Six Models Deprecated December 14, What to Fix

Six GitHub Copilot models are deprecated on December 14, 2026, and the official November 13, 2026 changelog gives six migration sentences but no configuration file name, model ID string or migration script. This SOP turns those sentences into a checkable list: replace hardcoded model names in workflows and integrations, enable model policies in Copilot settings, and understand what global default, the model selector, Copilot Chat, inline edits and code completions each do. All six model names and suggested alternatives are quoted as published, Kimi K3 stays To be announced, and the Claude Haiku 4.5 annual-plan exception for individual subscribers is spelled out.

Nov 16, 20269 min read
Field SOP

Bedrock AgentCore V2 Snapshots: Four Patterns, One Pre-warm

Turns the four official AWS section names into a checkable sequence for Bedrock AgentCore runtime V2 session startup: measure Latency and OverheadLatency in CloudWatch first, move imports and static data into init, use Working with hooks for build-time large files, then send a background warm-up request with Pre-warm on user engagement and keep the session ID. It records the 120-second /ping gate, that the V2 default applies only to new and updated runtimes, that existing ones need explicit platformVersion=V2, and that the sample code is for non-production use. The Region and cold-start discrepancies between the two official pages are kept side by side.

Nov 15, 20269 min read