News in Short
- Google has announced Gemini 4 Argon, a model focused on engineering and enterprise tasks.
- It can generate up to 1 million output tokens, up from 64,000 previously.
- Argon can work on software engineering, cybersecurity and business automation.
- Google says it can find, validate and patch software vulnerabilities.
- The model has been used internally for engineering projects at Google.
Google has introduced Gemini 4 Argon with a focus on complex software engineering, cybersecurity and enterprise tasks.
One of its biggest changes is the output limit. Gemini 4 Argon can generate up to 1 million output tokens. The previous limit stood at 64,000 tokens.
That larger output window could allow Argon to handle substantially longer coding and technical tasks in a single interaction.
Google has also tested the model on software engineering, long-video understanding and business automation benchmarks.
What can Gemini 4 Argon do for software engineering?
Software development is one of the main areas Google is targeting with Argon.
Google says the model scored 77.9 percent on the DeepSWE v1.1 software engineering benchmark.
The company has also used Argon internally for engineering projects. These include work involving quantum computing and large-scale codebase migrations.
In one project, Argon helped Google develop a Rust version of the libgav1 video decoder.
According to Google, the resulting decoder runs 2.7 times faster than the existing Rust implementation. It also produces the same video output.
The model’s capabilities therefore extend beyond generating code snippets. Google is positioning Argon for larger engineering workflows and complex codebases.
What can Gemini 4 Argon do for cybersecurity?
Cybersecurity is another major focus for the new model.
Google says Gemini 4 Argon can find, validate and patch software vulnerabilities. It scored 68 percent on CWE-bench v1, where Google says it tied for the top score.
The company is initially giving cybersecurity professionals access through its Fairwind Program.
During this testing phase, Google says Argon will operate without its usual cyber guardrails. This allows cybersecurity professionals to evaluate its capabilities more directly.
At the same time, Google is testing safeguards designed to reduce potential cyber misuse.
What can Gemini 4 Argon do with long-form tasks?
The expanded output limit is another major capability.
Argon can produce up to 1 million output tokens. That is more than 15 times the previous 64,000-token limit.
This could make the model more suitable for lengthy technical workflows. These could include large code migrations, detailed software analysis and other tasks that require extensive outputs.
Google also says Argon scored 91.7 percent on LVBench for long-video understanding.
However, the supplied information does not specify how the model achieved that score or what practical consumer features will use the capability.
What can Gemini 4 Argon do for businesses?
Google is also positioning Argon for business automation.
The model scored 51.3 percent on Zapier’s AutomationBench, according to Google.
The company has not yet detailed specific consumer-facing automation features. Instead, the initial focus is on developers, enterprises and professional users.
Argon is priced at $2 per million input tokens and $10 per million output tokens.
When will Gemini 4 Argon be available?
Google is initially testing Gemini 4 Argon through its Fairwind Program with selected cybersecurity professionals.
The company plans to expand access to paid API customers and Google AI Ultra subscribers. Developers, enterprises and consumers are expected to receive access later.
Google is also testing safeguards against cyber and CBRN misuse, indirect prompt injection and unintended model behaviour.
These tests will take place before wider availability.
For now, Gemini 4 Argon’s positioning is clear: Google is targeting longer, more complex technical and enterprise workloads rather than presenting it primarily as a general-purpose consumer chatbot.
