Standard (introductory)
$2 input / $10 outputper 1M tokens, through November 30, 2026Cached input $0.10 per 1M tokens via implicit caching. Output includes thinking tokens. Official source 1 ↗ · Official source 2 ↗
Reviewed tool profile
Gemini 4 Argon
Google’s flagship Gemini 4 reasoning model for hard coding, agent and research tasks, with a 1M-token window.
Sources checked
- Pricing
- Paid Gemini API only, billed per million tokens, with output price including thinking tokens. Standard pricing is introductory through November 30, 2026: $2 input, $10 output and $0.10 cached input. From December 1, 2026 it doubles to $4 input, $20 output and $0.20 cached input. Batch and Flex cost half of Standard, and Priority costs 1.8 times Standard. Grounding with Google Search is free for 5,000 requests a month, then $14 per 1,000. Official source 1 ↗ · Official source 2 ↗
- Available as
- Gemini API (model ID gemini-4-argon) and Google AI Studio on the paid tier, plus Android Studio and Google Antigravity. · Gemini app for Google AI Pro and Ultra subscribers, including Antigravity’s agentic coding interface in the Gemini macOS app. · Google also lists Gemini for business, Google AI Mode and Gemini Enterprise Agent Platform for enterprise deployments. Official source 1 ↗ · Official source 2 ↗ Official source 1 ↗ Official source 1 ↗ · Official source 2 ↗
- Primary group
- Coding & Development
What it does
Gemini 4 Argon is Google’s flagship reasoning model and the first model in the Gemini 4 family, announced on September 30, 2026 and opened to paid Gemini API developers and Google AI Pro and Ultra subscribers on October 13, 2026. It is built for complex software engineering, long multi-step agents, math and science, and enterprise work such as legal and financial analysis. It takes text, image, video, audio and PDF input, returns text, shares one 1M-token window between input, thinking and output, and can generate up to 1M tokens in one task through long decode.
Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗ · Official source 4 ↗ · Official source 5 ↗Features
- Unified 1M-token window. Input, thinking and output share one 1,048,576-token window. Official source 1 ↗ · Official source 2 ↗
- Long decode up to 1M tokens. Generates past the 64K streaming and 32K non-streaming per-request caps, up to 1M tokens minus the input; the latest Google GenAI SDKs continue automatically. Official source 1 ↗
- Three thinking levels. thinking_level low, medium (default) or high trades quality against latency and token use. Official source 1 ↗
- Built-in tools. Function calling, structured outputs, code execution, file search, URL context, Google Search and Maps grounding, and computer use in preview. Official source 1 ↗
- Multimodal input. Accepts text, images, video, audio and PDF and returns text; agentic video understanding is generally available. Official source 1 ↗ · Official source 2 ↗
- Agentic coding and migrations. Google reports 77.9% on DeepSWE v1.1 and uses Argon agents to migrate C/C++ codebases to Rust. The Antigravity agent in Gemini Managed Agents supports the model. Official source 1 ↗ · Official source 2 ↗
- Cybersecurity defense. Google says Argon can find, validate and patch software vulnerabilities; a version without cyber guardrails goes to trusted defenders through the Fairwind Program. Official source 1 ↗
- Batch, Flex and Priority processing. Batch and Flex at half the Standard price, Priority at a premium for lower latency. Official source 1 ↗ · Official source 2 ↗
Pricing & free limits
Standard (from December 1, 2026)
$4 input / $20 outputper 1M tokensCached input $0.20 per 1M tokens. Official source 1 ↗ · Official source 2 ↗
Batch
$1 input / $5 outputper 1M tokens (introductory); $2 / $10 from December 1, 2026Half the Standard price for asynchronous jobs. Official source 1 ↗
Flex
$1 input / $5 outputper 1M tokens (introductory); $2 / $10 from December 1, 2026Half the Standard price for latency-tolerant work. Official source 1 ↗
Priority
$3.60 input / $18 outputper 1M tokens (introductory); $7.20 / $36 from December 1, 2026Premium tier for higher-priority processing. Official source 1 ↗
Google AI Ultra (AI Studio)
Google AI Ultra subscriptionsubscriptionUltra subscribers can use Gemini 4 Argon in AI Studio with the highest prototyping limits; the AI Pro plan does not include it in AI Studio. Official source 1 ↗
Billing details
Output pricing includes thinking tokens, and thinking tokens count toward max_output_tokens. Official source 1 ↗
Only implicit caching is supported: it applies to prompts of at least 4,096 tokens with a 95% discount on cache hits. Explicit caching returns an error. Official source 1 ↗
Paid-tier content is not used to improve Google’s products. Official source 1 ↗
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Long-horizon software engineering, such as large codebase migrations and the hardest coding or agent tasks. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗
Enterprise knowledge work and document analysis, including legal and financial research. Official source 1 ↗ · Official source 2 ↗
Finding and patching software vulnerabilities for cyber defense teams. Official source 1 ↗
Generating very long outputs, such as full proofs, reports or code, in one task with long decode. Official source 1 ↗
- No free tier. Not available on the Gemini API Free tier or in AI Studio on the Google AI Pro plan. Official source 1 ↗ · Official source 2 ↗
- Text output only. Audio generation, image generation and the Live API are not supported. Official source 1 ↗
- Lower per-request cap. Non-streaming requests return at most 32,768 tokens, down from 65,536 on Gemini 3.x; longer output needs long decode. Official source 1 ↗
- API migration changes. Explicit caching is unsupported, sampling parameters such as temperature are ignored, and Google advises client timeouts of at least 20 minutes. Official source 1 ↗
- Knowledge cutoff and reliability. Knowledge cutoff is March 2026; Google notes possible hallucinations and occasional slowness or timeouts. Official source 1 ↗
- Introductory price ends. Prices double on December 1, 2026. Official source 1 ↗
Editorial
Featured in
Shortlists, alternatives and comparisons that include Gemini 4 Argon.
Similar tools
Tools like Gemini 4 Argon
Published tools that share its categories.
Claude Opus 5.5Anthropic’s Opus-tier Claude model for long-running agentic coding and knowledge work, at $4/$20 per million tokens.
Visit 
ChatGPTOpenAI's AI assistant for chat, search, files, images, coding with Codex and agent work across your apps.
Visit 
DeepSeekFree AI chat app with web search, DeepThink reasoning and file upload, plus a pay-as-you-go API for DeepSeek's open-source language models.
Visit 
GeminiGoogle's AI assistant for chat, document summaries, Deep Research reports, writing, coding, and image, video and music generation.
Visit 
GPT-6 LunaOpenAI’s cheapest GPT-6 model for focused, high-volume API work, with a 1.05M-token context.
Visit 
Claude Haiku 5.5Anthropic’s fastest, lowest-cost Claude model for high-volume tasks and coding subagents.
Visit