Google has moved into a new Gemini generation. On September 30, 2026, it announced Gemini 4 Argon, a frontier model built for long, multi-step work across software engineering, finance, legal research and cybersecurity. “Gemini 4.0” is already being used online, but Google’s official name is Gemini 4 Argon.
The headline change is not just better benchmark scores. Gemini 3.1 Pro had a 1 million-token input context window and a 64K output limit. Argon raises output to 1 million tokens, giving it far more room to continue difficult tasks in one trajectory. Access is also unusually restricted: trusted cyber defenders are getting it first through Google’s Fairwind Program.
Key Highlights
- Gemini 4 Argon was announced on September 30, 2026.
- Output rises to 1 million tokens, versus 64K on Gemini 3.1 Pro.
- Google reports 77.9% on DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on LVBench.
- Wider access is expected to begin with paid API customers and Google AI Ultra subscribers.
Gemini 4 Argon vs Gemini 3: The Biggest Technical Changes
Gemini 3 launched in November 2025 with stronger multimodal reasoning, agentic coding and a 1 million-token context window. It later spread across Search, the Gemini app, AI Studio, Vertex AI and Google Antigravity. Google’s original Gemini 3 launch details
Argon shifts the focus toward staying useful throughout longer jobs.
Three differences stand out:
- Longer output: Argon can produce up to 1 million output tokens instead of the 64K limit listed for Gemini 3.1 Pro.
- Long-horizon agents: It is built to reason, use tools and execute steps across extended workflows.
- Cyber defense: Google trained Argon to find, validate and patch software vulnerabilities autonomously.
That security capability is also shaping how Google is releasing the model.
Why Google Is Releasing Argon To Cyber Defenders First
Argon is entering the Fairwind Program before a broad consumer launch. Google says trusted defenders can use its full cybersecurity capability while the company keeps testing safeguards against misuse and prompt injection.
Google also points to an early case involving Wiz. Through its Scan for Good initiative, Argon reportedly found a critical vulnerability affecting healthcare software that previous frontier models had missed.
On CWE-bench v1, which measures vulnerability remediation, Argon scored 68%. DeepSWE v1.1, focused on long-horizon software engineering, puts it at 77.9%, according to Google.
Google DeepMind’s official X announcement also confirmed the rollout, describing Argon as a model for coding, enterprise knowledge work and cybersecurity defence.
Gemini 4 Argon Benchmarks Point To Longer AI Workflows
Argon’s launch figures are aimed heavily at developers and businesses. Google reports 51.3% on AutomationBench, a Zapier benchmark for completing end-to-end business tasks, and 91.7% on LVBench for long-video comprehension.
Google says Argon agents identified memory optimisations that freed more than 300 TiB across its data-centre fleet. Another project involved the libgav1 video decoder, where Argon replaced 32,000 lines of SIMD code and produced a Rust version that Google says ran 2.7 times faster than the previous port.
For users, that points toward four practical workloads:
- Large code migrations running across many steps.
- Legal or financial research across several documents.
- Video, chart and document analysis inside one workflow.
- Agents that act, inspect results and continue rather than stopping after one answer.
The comparison is therefore less about a prettier response and more about how long the model can keep working coherently.
When Can Users Get Gemini 4 Argon And What Will It Cost?
Google has not given a date for broad public access. It plans to gather feedback from trusted testers before expanding availability to developers, enterprises and consumers. Paid API users and Google AI Ultra subscribers are expected among the first wider groups.
Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input priced at a 95% discount. Google says the rates will later rise to $4 and $20.
Gemini 4 Argon is therefore not simply Gemini 3 with a new label. Its biggest change is endurance: much longer output, deeper agentic work and a stronger cybersecurity focus. Developers and businesses will get the clearest answer once broader access begins and those benchmark claims meet everyday workloads.
Follow Google DeepMind and official Gemini updates before changing workflows or budgets, as access, pricing and model availability are still moving quickly.
FAQs
Is Gemini 4 Argon the same as Gemini 4.0?
Gemini 4 Argon is Google’s official name. “Gemini 4.0” is informal shorthand being used online for the new generation, rather than official product branding.
How is Gemini 4 Argon different from Gemini 3?
Argon targets longer workflows, coding, enterprise work and cyber defence, while lifting maximum output from 64K tokens on Gemini 3.1 Pro to 1 million.
Can anyone use Gemini 4
No. Google is initially limiting access to selected cyber defenders through Fairwind, with paid API customers and Google AI Ultra subscribers expected later.
What benchmark scores has Gemini 4 Argon posted?
Google reports 77.9% on DeepSWE v1.1, 51.3% on AutomationBench and 91.7% on LVBench, alongside a 68% result on CWE-bench v1.
How much will Gemini 4 Argon cost?
Introductory API pricing is $2 per million input tokens and $10 per million output tokens. Google says standard pricing will later double those rates.
Related Tech Picks: AI, Google & Smart Devices
What’s New in Gemini 3.7 Flash?
Check out the latest changes in Google’s faster AI model.
Can Indian Students Get AI Plus?
Find out about Google’s one-year AI Plus student offer.
Can Gemini AI Run Apps?
Uncover how Gemini intelligence works across Samsung foldables.
How Can Students Access Gemini?
Explore Google’s AI Plus offer with Gemini Omni and storage.
What Does Google Fitbit Air India Offer?
Dive into its price, battery life, health features, and availability.
