Long-horizon reasoning
The model is designed to sustain reasoning through complex, long-horizon workflows rather than limiting work to short, isolated exchanges.
Gemini 4 Argon
Gemini 4 Argon is Google’s frontier AI model for long-horizon software engineering, enterprise knowledge work and cybersecurity defense. It is initially rolling out to trusted cyber defenders, with broader access planned after further testing and guardrail development.
What is Gemini 4 Argon?
Gemini 4 Argon is Google’s frontier AI model for complex, long-horizon professional workflows. It combines advanced reasoning, coding, multimodal understanding and cybersecurity capabilities for work such as software engineering, financial research, legal drafting and vulnerability remediation.
The model is being introduced gradually. It is initially rolling out to trusted cyber defenders through Google’s Fairwind Program while Google continues pre-release testing, gathers feedback and refines safeguards before expanding access to developers, enterprises and consumers.
Core capabilities
Software engineering
Argon supports everyday debugging, algorithm design, large-scale codebase migrations and optimization. Google reports a 77.9% result on the DeepSWE v1.1 real-world software-engineering benchmark.
Enterprise knowledge work
The model is positioned for multi-step financial research, legal research and drafting, tax-related work and broader enterprise knowledge tasks. Google reports leading results on the Vals Index, Vals Finance Agent v2 and Harvey’s Legal Agent Benchmark.
Multimodal analysis
Argon can interpret professional charts, extract details from long videos and use information distributed across a series of documents. Google reports a 91.7% score on LVBench for long-video understanding.
Cybersecurity defense
For trusted defenders and Google's internal teams, Argon is being released without cyber guardrails to support autonomous vulnerability discovery, validation and patching. It ties for first on CWE-bench v1 with a reported score of 68%.
Extended output capacity
The model has a stated 1-million-token output limit, up from 64,000 tokens, giving a single task trajectory more room for extended analysis and generated work.
Practical use cases
Large-scale software engineering
Engineering teams can use Argon for debugging, algorithm design, profile-guided optimization and migrations of large C/C++ codebases to Rust. The source describes work ranging from core libraries to the Fuchsia Zircon kernel, with auditing and testing before production rollout.
Research and document-heavy knowledge work
Finance, legal and tax teams can apply the model to multi-step research, document-based analysis, legal research and drafting. Its long output capacity is intended for tasks that require sustained reasoning rather than a single short answer.
Visual and video analysis
Organizations can use its multimodal abilities to analyze professional charts, inspect long videos and combine evidence across multiple documents when making or preparing business decisions.
Vulnerability discovery and remediation
Trusted cyber defenders can use Argon to find, validate and patch critical software vulnerabilities. Google says the model is being made available without cyber guardrails to trusted defenders and internal teams for this defensive purpose.
Technical research and infrastructure optimization
Research and infrastructure teams can assign the model extended technical investigations, such as optimizing quantum-computing subroutines or analyzing fleet-wide profiling telemetry for memory improvements.
Pros and Cons
Pros
- Supports extended, multi-step workflows with a stated 1-million-token output limit.
- Covers software engineering and enterprise knowledge work, including finance and legal tasks.
- Adds specialized capabilities for finding, validating and patching software vulnerabilities.
- Handles multimodal work involving charts, long videos and multiple documents.
- Google reports concrete internal engineering results, including a 2.7x faster Rust video decoder than an existing Rust port in one libgav1 project.
Cons
- General availability is not yet specified; the initial rollout is limited to trusted cyber defenders and other access details are not provided.
- The announcement does not document supported platforms, APIs, integrations or deployment surfaces.
- The reported benchmark results and internal Google examples come from Google or partner announcements and are not independent evaluations in the supplied source.
FAQ
Who can access Gemini 4 Argon?
Google says Gemini 4 Argon is rolling out first to trusted cyber defenders through the Fairwind Program, with broader access for developers, enterprises and consumers planned after further testing and guardrail work. The source does not specify a general availability date.
What is the announced pricing for Gemini 4 Argon?
The announced introductory price is $2 per million input tokens and $10 per million output tokens. Cached input tokens are priced at 95% below the input-token price. The source does not provide further plan or quota details.
What kinds of work is Gemini 4 Argon designed for?
Argon is intended for long, multi-step workflows including software engineering, financial research, legal research and drafting, visual document and chart analysis, and cybersecurity vulnerability discovery and patching.
How large is Gemini 4 Argon's context or output capacity?
The model has a stated output-token limit of 1 million tokens, increased from the previous 64,000-token limit. This is intended to support extended reasoning and large, complex task trajectories.
Quick Facts
- Category
- Frontier AI model
- Developer
- Google DeepMind / Google
- Primary workflows
- Coding, enterprise knowledge work and cybersecurity defense
- Output limit
- Up to 1 million tokens
- Initial access
- Trusted cyber defenders through the Fairwind Program
- Announced introductory price
- $2 per million input tokens; $10 per million output tokens
Альтернативы Gemini 4 Argon
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.
hob
hob is an independent workspace for coding agents that keeps agent sessions, terminals, history, and follow-up work organized around the tools and providers you already use. It is aimed at developers who want local control over routing, history, and workspace structure rather than a bundled model stack.
Grok 4.7
Grok 4.7 is an xAI model for software development and knowledge work, including extended coding tasks and professional analysis. It is available through Grok Build, Cursor, the Grok API, and other developer platforms, with token pricing starting at $2 per million input tokens and $6 per million output tokens.
Manta AI
Manta AI is an autonomous web app testing tool for teams that want to map application behavior, catch regressions, and generate tests without writing scripts or maintaining selectors. It works from a URL and supports plain-English test flows, run results with screenshots, and scheduled or deployment-triggered checks.