Gemini 4 Carbon: what Google announced and what you cannot use yet

1. What Google announced#

On October 28, 2026 Google introduced Gemini 4 Carbon as its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It arrives two weeks after Gemini 4 Argon, on which it builds. The table summarizes what Google published and, above all, what it did not.

One key detail: the announcement does not mention the API, Google AI Studio or the Gemini app. Today there is no public way to call the model.

Documentation: Google: introducing Gemini 4 Carbon ↗

2. Who can use it and when it reaches everyone else#

According to Google, access starts that same week with a subset of cyber defenders and expands gradually to more trusted partners through the Fairwind Program. SiliconANGLE describes it as a program giving early access to organizations, for example internet providers, that use the models to find and fix security flaws in their systems.

General release to developers, businesses and consumers is «as soon as possible», with no date. Google says it will keep refining the guardrails with early users' feedback, and that it takes part in the U.S. government's voluntary pre-release testing process.

Documentation: SiliconANGLE: Google debuts Gemini 4 Carbon ↗

3. What it will cost#

Google states an introductory rate of $2 per million input tokens and $10 output, with 95% off input served from cache, and $4 and $20 after the introductory period. These are the same figures as Gemini 4 Argon. The announcement does not say when Carbon's introductory period ends: Argon's ends on November 30, 2026.

The code calculates an example month (150 million input tokens and 30 million output tokens) with and without caching. It is for planning only: you cannot buy Carbon yet and how caching will apply is not described.

from decimal import Decimal as D

RATES = {"until the introductory rate ends": (D("2"), D("10")), "afterwards": (D("4"), D("20"))}   # USD per million tokens
CACHE_DISCOUNT = D("0.95")             # 95% off input served from cache
TRM = D("3148.02")                      # TRM of Oct 28, 2026
INPUT_M, OUTPUT_M = D(150), D(30)     # an example month: 150 M input and 30 M output

def month(rate, cached_fraction):
    t_in, t_out = rate
    input_cost = INPUT_M * ((1 - cached_fraction) * t_in + cached_fraction * t_in * (1 - CACHE_DISCOUNT))
    return input_cost + OUTPUT_M * t_out

for name, rate in RATES.items():
    for fraction in (D("0"), D("0.6")):
        usd = month(rate, fraction)
        print(f"{name}, {fraction:.0%} cached: ${usd:,.2f} | COP {usd * TRM:,.0f}")

Documentation: Google: introducing Gemini 4 Argon ↗

4. What Google claims#

The table gathers the results Google publishes. They are the company's own figures, and Google clarifies that the security tests were run without guardrails. None has been independently verified in the sources we consulted.

5. What to do as a developer today#

• Do not change your architecture for a model you cannot call: keep going with Gemini 4 Argon and its price dates.
• Your budget does not change because of Carbon: it has the same rates as Argon, which double when each one's introductory period ends.
• If you work in cybersecurity (vulnerability detection, patching), look at the Fairwind Program requirements: it is the access route today.
• Prepare an evaluation with your own tasks for when there is general access: an announcement's results rarely match your case.
• Note the unknowns to check on launch day: model id, context window, API availability and the date the introductory price ends.

6. What we do not know#

• The model id and its context window.
• Whether there will be access through the Gemini API, AI Studio or the enterprise platform, and on what date.
• When Carbon's introductory price ends.
• How the guardrails will apply to someone using the model for legitimate offensive security.
• Independent results: the figures in the table are Google's.

7. Keep your numbers current#

This guide is based on Google's October 28, 2026 announcement. Availability, prices and conditions may change when it leaves restricted access; check the official Gemini API documentation.

• Check Google's blog and the Gemini API documentation.
• Recalculate in pesos with the TRM of the day you budget.
• Note the reading date of each rate.