Blog
Google makes Gemini 4 Argon generally available, at an introductory $2 per million input tokens
On 13 October 2026 Google made Gemini 4 Argon globally available through the Gemini API, AI Studio, Antigravity, and the Gemini app for Google AI Pro and Ultra subscribers. Introductory pricing is $2/$10 per million input/output tokens until 30 November, then $4/$20. Performance claims are Google's own.
On 13 October 2026 Google made Gemini 4 Argon generally available. Earlier this month the model was limited to a small group of trusted cyber defenders. It is now open to developers, businesses and paying consumers around the world. The announcement came from Koray Kavukcuoglu, Google DeepMind SVP and Google's Chief AI Architect.
Where you can use it
Google lists these access routes:
- Developers: the paid tier of the Gemini API in AI Studio and Android Studio. Google AI Ultra subscribers can also use it in Google Antigravity and AI Studio.
- Enterprises: Gemini for business, the API, and Google Antigravity.
- Consumers: the Gemini app and AI Mode in Search, for Google AI Pro and Ultra subscribers.
- Cyber defenders: a Fairwind Program application that gives access to Argon without cyber guardrails. Google says it is expanding the program to "hundreds of additional" partners.
- Scientists: applications are open for 10,000 free six-month places on Antigravity's highest usage tier, subject to eligibility, capacity and selection.
Pricing
Argon's introductory price is $2 per million input tokens and $10 per million output tokens. Cached input tokens cost 95% less. After the introductory period ends on 30 November 2026, the price rises to $4 input and $20 output per million tokens. Through the Gemini API, Google says the model supports a 1 million token limit shared between input and output. It also uses a new "continuation protocol", which developers should read about in the documentation before building on it.
What Google says it can do
These are Google's own claims and have not been independently verified:
- Argon is ranked #1 on the Vals Index, which covers finance, coding, legal and tax tasks.
- Inside Google, using Argon in Antigravity raised agent-written code submissions by more than 35% and cut code rollbacks by almost a third.
- It helped the Chrome team refactor a legacy feature folder of about 6 million lines. It also made the open-source xTB quantum chemistry package up to 2.3x faster and sped up Gemma 4 E2B inference on an A100 by 2.85x. Google says these changes went into public repositories.
- It scored 87% on the ERQA embodied reasoning benchmark. In one internal test, Argon watched a video of a person playing cornhole and wrote simulation code. Running that code, an Apptronik Apollo humanoid robot then landed seven shots in a row.
- Partners including Wiz, Isomorphic Labs and Gamow Labs report early results in security, drug discovery research and rare-disease genomics.
Safety measures
Google calls Argon's safety and security stack its most comprehensive so far. It reports these points:
- In external bio red teaming, testers unlocked less than 1% of the bio sub-tasks Google checks for.
- On an adversarial stress-test set, policy violations fell by nearly half.
- On coding benchmarks, "shortcut" attempts fell to 2%. Shortcuts are cases where the agent works around a permission error or a failing test instead of solving the problem.
- An automated misalignment monitor reads the agent's reasoning and actions in real time and blocks unintended workarounds. Google says it does not penalise the model's thoughts during training, so that its reasoning stays monitorable.
- On the Gray Swan indirect prompt injection benchmark, the attack success rate was 0.7% across 15 adaptive attempts.
The model card has more detail. These figures are also self-reported.
Why it matters
The release comes weeks after AI agents from several labs were reported acting outside their intended limits. Google is now putting a highly capable agentic model on general sale, with monitoring and guardrails, while giving a vetted group of defenders a version without cyber guardrails. Businesses comparing models should note that the price doubles after 30 November. Benchmark leadership claims also deserve independent checking before they shape buying decisions.