Google has announced Gemini 4 Argon, the first model in the Gemini 4 family. It is built for software engineering, legal and finance work, and cybersecurity defense. Access is restricted for now, so most developers and Google AI subscribers cannot use it yet.
Google revealed the model on September 30 in a blog post from Koray Kavukcuoglu, SVP of Google DeepMind. Gemini 4 Argon is rolling out to a set of trusted cyber defenders through the Fairwind Program. Google is also taking part in the US government’s voluntary process for pre-release model access while it expands availability in stages.
Fairwind Partners Get Gemini 4 Argon Without Cyber Guardrails
Google trained Gemini 4 Argon to find, validate and patch critical software vulnerabilities on its own. Trusted defenders and Google’s internal teams are getting the model without cyber guardrails, so they can use its full defensive capability.
Wiz is already running Gemini 4 Argon through its Scan for Good program, which scans critical public infrastructure for free. Google says the model found a critical flaw in healthcare software used by hospitals worldwide, exposing sensitive personal information, a risk earlier frontier models had missed.
Google says the model is also its most resilient yet against indirect prompt injection. For the wider release, it is adding monitoring of the model’s internal activations to catch misuse, plus checks on its chain-of-thought and actions that stop execution when Gemini 4 Argon goes beyond what the user intended.
Google Reports 77.9% on DeepSWE v1.1 and Has Argon Rewriting C and C++ in Rust
These are the benchmark scores Google published for Gemini 4 Argon, and all of them are Google’s own figures.
- DeepSWE v1.1 (long-horizon software engineering) at 77.9%, which Google calls a new state of the art
- AutomationBench from Zapier at 51.3%, ranked first
- LVBench (long video understanding) at 91.7%, also a state of the art result
- CWE-bench v1 (fixing security vulnerabilities) at 68%, tied for first place
Google also says Gemini 4 Argon leads the Vals Index, Vals Finance Agent v2 and Harvey’s Legal Agent Benchmark.
Inside Google, Argon agents are migrating C and C++ codebases to Rust, from core libraries like re2 and libgav1 up to more than 800,000 lines for the Fuchsia Zircon kernel. In libgav1, the open source video decoder, Argon replaced 32,000 lines of SIMD code and produced a memory-safe decoder that runs 2.7x faster than the existing Rust port with identical video output. Separately, a team of Argon agents applied memory optimizations across Google’s data centers that free up more than 300 TiB once rolled out.
Gemini 4 Argon Starts at $2 Per Million Input Tokens and Doubles Later
Google set an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input price. Output limit goes up to 1 million tokens, from 64,000 on the previous generation. Google released Gemini 3.7 Flash in August at an introductory $0.75 per million input tokens and $3.75 per million output tokens.
Wider access starts with paid API customers and Google AI Ultra subscribers. Google has not given a date. Once the introductory period ends, the price moves to $4 per million input tokens and $20 per million output tokens.
