Software Engineering Collective

News · SI

Gemini 4 Argon is now generally available to developers

Google has made Gemini 4 Argon, its new frontier model for long-horizon coding and agent work, available through the Gemini API at an introductory $2 and $10 per million tokens. The price doubles after 30 November 2026, so the time to benchmark it is now.

Spectrum Web Co3 min read

GoogleGeminiCoding agentsPricing

Google has made Gemini 4 Argon, its newest frontier model, globally available to developers and enterprises from 13 October 2026 (US time), at an introductory API price that doubles after 30 November 2026[^1].

01 / 04

What shipped

Argon is available on the paid tier of the Gemini API in Google AI Studio and Android Studio[^2]. Google AI Ultra subscribers can also use it in AI Studio and in Google Antigravity, and enterprises can use it in Gemini for business, the API and Antigravity[^1]. Consumers on Google AI Ultra and Pro get it in the Gemini app and AI Mode in Search[^1].

Google says Argon is built for long-horizon work across coding, finance, legal, cybersecurity and science, and that it ranks first on the Vals Index[^1].

02 / 04

Pricing and limits

  • Introductory price: $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off[^1].
  • After 30 November 2026: $4 per million input tokens and $20 per million output tokens[^1].
  • A 1 million token unified input-output limit via the Gemini API, with a new continuation protocol for long tasks[^1].

03 / 04

What Google says it can do

  • Argon in Antigravity helped Google teams lift agentic code submissions by more than 35% while reducing code rollbacks by nearly a third[^1].
  • Working overnight on Google's open-source Agent Development Kit, it found optimisations that cut in-memory session prep time by 40% in long sessions and sped up cold imports by 21%[^1].
  • It delivered a 2.85x end-to-end inference speedup for Gemma 4 E2B on an A100 via vLLM, from about 37,900 to over 108,000 tokens per second, with 100% token parity[^1].
  • It scored 87% on the Embodied Reasoning (ERQA) benchmark for physical and spatial problems[^1].

On safety, Google reports shortcut attempts on coding benchmarks dropped to 2%, and attack success on the Gray Swan Indirect Prompt Injection benchmark fell to 0.7% across 15 adaptive attempts[^1]. All of these are Google's own figures.

04 / 04

Spectrum take: what to do before 30 November

The headline for shipping teams is the pricing cliff. Until 30 November 2026 Argon costs $2 and $10 per million tokens; after that it is $4 and $20. That gives you about six weeks to benchmark it cheaply, and a clear reason to model your costs at the higher rate, not the launch rate.

What we would test first:

  • Your hardest refactor or migration. Argon is pitched at long, multi-file work. Run it against a real ticket in a branch and measure merged changes and rollbacks, the same two numbers Google reports.
  • Long agent sessions. The 1M unified input-output limit and continuation protocol change how long tasks are chunked. Read the docs before porting an agent, since output now draws on the same budget as input.
  • Caching. With cached input 95% off, agents that resend the same repo context or system prompt should restructure prompts so the stable prefix is cached. That matters even more once the price doubles.
  • Prompt injection in your own stack. A 0.7% benchmark figure is encouraging, but agents that read email, web pages or tickets still need tool permissions and human approval on irreversible actions.

If you already run a cheaper model for high-volume subagents, keep it. Argon makes most sense as the planner or the model you hand the hard ticket to, with costs checked against the post-November price.

FAQ

Questions, answered.

01

What is Gemini 4 Argon?

Gemini 4 Argon is Google's latest frontier model, announced on 13 October 2026 (US time). Google positions it for long-horizon, multi-step work such as large codebase refactors, code migrations, cybersecurity and scientific research. It is the first model of the Gemini 4 family to ship.

02

How much does Gemini 4 Argon cost?

Through the Gemini API it launches at an introductory $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off. After the introductory period ends on 30 November 2026, pricing rises to $4 per million input tokens and $20 per million output tokens.

03

Where can developers use Gemini 4 Argon?

Developers can build with it on the paid tier of the Gemini API in Google AI Studio and Android Studio. Google AI Ultra subscribers can also use it in AI Studio and in Google Antigravity, Google's agent-first development platform. Enterprises can use it in Gemini for business, the API and Antigravity.

04

What is the Gemini 4 Argon context limit?

Google says Argon has a 1 million token unified input-output limit via the Gemini API. It is paired with a new continuation protocol, and Google points developers to its documentation to set it up correctly for long-running tasks.

05

Is Gemini 4 Argon good at coding?

Google reports that Argon in Antigravity helped its teams increase agentic code submissions by more than 35% while cutting code rollbacks by nearly a third. It also cites a 40% cut in session prep time in Google's Agent Development Kit and a 2.85x vLLM inference speedup on Gemma 4 E2B. These are vendor figures, so test on your own repositories.

06

How does Gemini 4 Argon handle prompt injection and agent safety?

Google says Argon leads the Gray Swan Indirect Prompt Injection benchmark, with attack success cut to 0.7% across 15 adaptive attempts. On coding benchmarks, shortcut attempts dropped to 2%, and the model is trained to pause for human confirmation before irreversible steps.