# Gemini 4 Argon, the first Gemini 4 model, is now available to developers and in the Gemini app

> On October 13, 2026, Google opened Gemini 4 Argon, the first model in the Gemini 4 family, to paid Gemini API customers and to Google AI Pro and Ultra subscribers in the Gemini app. Argon is built for long-running work in coding, legal and finance, and cybersecurity, and can output up to 1 million tokens per request. It's coming to AI2Face soon.

Published 2026-10-14 · Updated 2026-10-14
URL: https://ai2face.com/news/gemini-4-argon

On September 30, 2026, Google DeepMind announced **Gemini 4 Argon**, the first model in the Gemini 4 family, and on October 13 it opened the model to developers and Gemini app users. Google calls Argon a frontier model for long, multi-step work: software engineering, legal and finance tasks, research and cyber defense.

## What is Gemini 4 Argon, the first Gemini 4 model?

According to the model card, Argon is a new model, not a fine-tune of an earlier Gemini. It's a large sparse mixture-of-experts transformer. The model understands text, images, audio, video and PDFs, and only outputs text. The context window is 1,048,576 tokens for input and output combined, and the knowledge cutoff is March 2026.

The headline technical change is output length. Previous Gemini models produced up to 64,000 tokens; Argon can generate up to 1 million tokens per request in long decode mode. Without that mode, the output limit is 65,536 tokens.

Google already uses Argon internally, and thousands of employees work with it every day. Argon-powered agents rewrote 32,000 lines of SIMD code in the libgav1 video decoder: the new decoder is memory-safe and runs 2.7x faster than the Rust port with identical output. Other agents found optimizations across Google's data centers that freed more than 300 TiB of memory, and they're now migrating the Zircon kernel from the Fuchsia OS to Rust — more than 800,000 lines of code.

## Gemini 4 Argon benchmarks: where it leads and where it trails

Google's table compares Argon with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5 across 19 benchmarks. Argon comes first in 13, ties for first in one and trails in five. The key rows:

| Benchmark | What it measures | Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index | legal, finance, coding, tax | **68.9%** | 63.1% | 67.0% |
| Harvey's Legal Agent Benchmark | legal tasks | **19.6%** | 5.4% | 3.8% |
| DeepSWE v1.1 | long-horizon software engineering | **77.9%** | 74.1% | 74.2% |
| GraphWalks 256K–1M | long-context reasoning | **84.2%** | 71.8% | 66.8% |
| LVBench | long video understanding | **91.7%** | 87.5% | 83.7% |
| CWE-bench v1 | vulnerability remediation | **68.0%** | **68.0%** | 67.0% |
| FrontierSWE v2 | hard software engineering | 55.0% | **65.5%** | 62.3% |
| Terminal-bench 4.0 | terminal tasks | 57.4% | 58.2% | **66.4%** |
| OSWorld-2.0 | computer use | 69.2% | **72.6%** | — |

Argon is strongest with documents and long context: on Harvey's legal benchmark, it solves nearly three times as many tasks as the closest rival. For agentic terminal work and the hardest software engineering tasks, GPT-6 Astra and Claude Opus 5.5 are still ahead.

Most of the numbers in the table come from Google's own runs. Independent results are close: Vals AI ranks Argon first on its index at 68.9%, and on the Artificial Analysis Intelligence Index v4.3.2, Argon with high reasoning scores 53 — eighth out of 216 models, level with GPT-6 Astra. Ahead of it are Claude Opus 5.5 (58) and [Meta's Muse Wonder Full](/news/muse-wonder) (56), released on October 12.

## Where is Gemini 4 Argon available and how do you get access?

- **Gemini API** — on the paid tier, through Google AI Studio and Android Studio. The model ID is `gemini-4-argon`. Google AI Ultra subscribers can try Argon in AI Studio and in Google Antigravity, Google's agent-first development platform.
- **Gemini app** — for Google AI Pro and Ultra subscribers. Alongside Argon, the chat gained a toggle for tasks mode: Gemini keeps working on a task even when you step away from your computer and connects to 40+ services, including Instacart, OpenTable, Adobe and Shopify.
- **Fairwind Program** — cyber defenders got Argon first, back in late September. Vetted defenders and Google's internal teams get a version without cyber guardrails: the model finds, validates and patches vulnerabilities in code on its own.

There's no free access to Argon, either in the API or in the app.

## How much does the Gemini 4 Argon API cost?

Google launched Argon at an introductory price and has already announced what it will cost afterward:

| | Introductory price | After the introductory period |
|---|---|---|
| Input, per 1M tokens | $2 | $4 |
| Cached input | $0.10 | $0.20 |
| Output, per 1M tokens | $10 | $20 |

Google hasn't said how long the introductory period will last. Argon uses a lot of output tokens: by Artificial Analysis's measurement, one task in its index costs $1.99 at the introductory price.

Argon has three thinking levels: low, medium and high. The model supports function calling, structured outputs, Google Search and Google Maps grounding, URL context, code execution and file search. Computer use is in preview. Argon can't generate images or audio.

## Gemini 4 Argon in AI2Face

Gemini 4 Argon is coming to AI2Face soon, together with text generation: the service is adding the top language models, including Google's new release. The price is shown before each generation.

Argon itself only outputs text. Images and video in AI2Face come from other Google models: [Nano Banana Pro](/models/nano-banana-pro) and [Nano Banana 2.1](/models/nano-banana-2-1) for images, and [Veo 3.1](/models/veo-3-1) and [Gemini Omni](/models/gemini-omni) for video.

## FAQ

### When was Gemini 4 Argon released?

Google announced the model on September 30, 2026, and on October 13 opened it in the paid Gemini API and in the Gemini app for Google AI Pro and Ultra subscribers. Before that, only cyber defenders in the Fairwind Program had access.

### How much does the Gemini 4 Argon API cost?

At the introductory price, it's $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. After the introductory period, prices rise to $4 and $20; Google hasn't said when.

### Is Gemini 4 Argon free?

No. In the Gemini API, the model is only on the paid tier, and in the Gemini app it's available to Google AI Pro and Ultra subscribers.

### Can Gemini 4 Argon generate images?

No. Argon accepts text, images, video, audio and PDFs, but only outputs text. Google's image models are Nano Banana; for video, it's Veo and Gemini Omni.

### Are Gemini 4 Pro and Gemini 4 Argon the same thing?

Google hasn't announced a model called Gemini 4 Pro. The first and so far only model in the Gemini 4 family is called Argon.

## Sources

- https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-4-argon/
- https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
- https://blog.google/innovation-and-ai/products/gemini-app/gemini-4-argon-availability/
- https://deepmind.google/models/model-cards/gemini-4-argon/
- https://ai.google.dev/gemini-api/docs/models/gemini-4-argon
- https://artificialanalysis.ai/models/gemini-4-argon
- https://www.vals.ai/models/google_gemini-4-argon