AIDEA Weekly
한국어

↑↓ Enter Esc

Google's strongest model, at the same launch price

GoogleGoogle's strongest model, at the same launch price

Google, Weekly 2026.10.25 – 10.31

Google's strongest model, at the same launch price

By Sangji, published , updated

On October 28, 2026, Google released Gemini 4 Carbon. The launch price is $2 input and $10 output per 1M tokens, the same as Argon; after the launch period it becomes $4 input and $20 output. For now it is open first to a limited group of cyber defenders.

Key facts

ReleasedOctober 28, 2026
Launch price$2 input, $10 output (per 1M tokens)
After launch period$4 input, $20 output
Cached input95% off the input price
Available now toA limited group of cyber defenders, Fairwind program
General availabilityNo date

What happened

Google released Gemini 4 Carbon on October 28, 2026, on its official blog. The model is aimed at cybersecurity and long-running agent tasks, and Google describes it as its strongest model so far.

Not open to everyone yet

From the week of the announcement it opened to a limited group of cyber defenders, and Google says it will expand to trusted partners through the Fairwind program. For developers, businesses and general users, Google said only that it will open as soon as possible after refining safeguards. There is no date. The post also does not describe how general developers can apply.

Pricing

The launch price is the same as Argon, released two weeks earlier. Cached input gets a 95% discount on the input price. By calculation, that is $0.10 per 1M tokens at the launch price. The length of the launch period was not given.

Gemini 4 Carbon pricing per 1M tokens. Input goes from $2 at launch to $4 after the launch period, and output from $10 to $20.
Price per 1M tokens. The end of the launch period was not given. Source: Google official blog

How it differs from Argon

  • Its success rate at generating secure code is more than twice Argon's. This is a Google internal benchmark.
  • Google says it cut unsupported claims by about half. This is also an internal measurement.
  • The Gray Swan indirect prompt attack success rate went from 0.7% to 0.0%.
  • It ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59.
  • It ranks fourth on WebDev Arena at 1768 Elo.

How to read the security scores

  • SEC-bench Pro (finding vulnerabilities) 88.2%, ExploitGym (attacks) 54.8%, CWE-bench (root-cause patching) 71.0%.
  • Google stated that these security scores were measured without safeguards.
  • Our reading: because its attack capability also rose, it is being opened to defenders first. Google did not explain it this way.

What you can do now

  • General users and developers have to wait for the release.
  • When budgeting, it is safer to plan around $4 input and $20 output rather than the launch price.
  • The post does not give the context size or rate limits.

Sources

Questions

How much does Gemini 4 Carbon cost?

The launch price is $2 input and $10 output per 1M tokens. After the launch period it is $4 input and $20 output.

Can I use it right now?

It is not generally available yet. Google says it is starting with a limited group of cyber defenders and expanding to trusted partners.

Can I trust the benchmarks?

They mix internal measurements and external rankings, and the security scores were measured without safeguards. This article notes which is which.