Gemini 4 Argon Pricing: Rate Card, Token Limits, and Budgeting

On September 30, 2026, Google DeepMind introduced Gemini 4 Argon, a frontier reasoning model designed for complex, long-horizon tasks across software engineering, enterprise knowledge work, and defensive cybersecurity. The release represents the vendor's newest foundation model architecture, engineered specifically to handle sustained multi-step workflows without losing coherence.

Unlike standard text models or earlier API endpoints capped at smaller completions, Gemini 4 Argon features a 1,048,576 unified input-output token limit per request, with output defaulting to 65,536 tokens and reaching up to 1,048,576 only with long decode. On standardized evaluations, the model records a score of 77.9% on DeepSWE v1.1 for real-world software engineering, leads GPT-6 Astra and Claude Opus 5.5 on Zapier's AutomationBench in Google's table at 51.3%, and ties for first place on CWE-bench v1 with a 68% remediation score.

For technical leads, operations directors, and compliance officers evaluating generative infrastructure, this release shifts the economics of running deep reasoning pipelines. Teams assessing Gemini 4 Argon pricing must evaluate whether the model's throughput and introductory token rates offset the expense of generating long-form trajectories in production.

API Rate Card: Input, Output, and Context Caching Costs

Gemini 4 Argon pricing opens at an introductory rate of $2.00 per million input tokens and $10.00 per million output tokens. For applications that leverage static context across repeated requests, cached input tokens receive a 95% discount, reducing cached prompt costs to $0.10 per million tokens.

Google DeepMind has structured this pricing to reward architectures that take advantage of prompt caching. Teams analyzing large code repositories, static legal corpora, or multi-document compliance libraries will see their heaviest prompt expenses occur only on the initial cache write.

Because rate cards frequently adjust after initial preview windows, operators must confirm current rates directly on Google DeepMind's official pricing portal before committing infrastructure budgets.

• Base input token pricing: $2.00 per million tokens.
• Base output token pricing: $10.00 per million tokens.
• Cached input token pricing: $0.10 per million tokens (95% discount off standard input).
• Unified context limit: 1,048,576 input-output tokens (output defaults to 65,536; up to 1,048,576 with long decode).

Gemini 4 Argon API Cost Per Call, Before and After December 1

A Gemini 4 Argon API call with 20,000 input tokens and 2,000 output tokens costs $0.06 at the introductory Standard rate, rising to $0.12 on December 1, 2026. At that volume, 1,000 identical calls cost $60 before December 1 and $120 once introductory pricing ends. Batch and Flex workloads cut those token charges in half, while Priority routing multiplies Standard tier rates by 1.8.

Google Search grounding includes 5,000 free requests each month, after which additional queries cost $14 per 1,000 requests. Google does not publish Argon's per-minute or per-day request and token limits; it points developers to AI Studio. You can compare those constraints across models in the full limits breakdown.

Availability: Developer Access and Consumer Plans

Google made Gemini 4 Argon generally available on October 13, 2026, across the paid Gemini API tier, Google Cloud, and Gemini for business. The Fairwind Program remains active specifically for vetted cybersecurity defenders requesting access without cyber guardrails.

Consumer access is available in the Gemini app and AI Mode for Google AI Ultra subscribers and rolling out to Google AI Pro subscribers in select countries.

Argon is not on the Gemini API free tier, so testing it requires a paid API key, a Google AI Ultra plan, or a Google AI Pro plan where Pro access has rolled out.

Comparing Argon Costs Against Prior Generations and Rival Models

At $2.00 per million input tokens and $10.00 per million output tokens, Gemini 4 Argon sits directly in the pricing bracket occupied by top-tier frontier models. It provides a more cost-effective input tier than many historic frontier launches, which frequently debuted at $5.00 to $15.00 per million input tokens.

The model's technical distinction lies in its 1,048,576 unified input-output token window. While output defaults to 65,536 tokens, enabling long decode allows completions to scale up to the full 1,048,576-token unified ceiling.

When evaluated against lightweight models such as Gemini 3.8 Flash Cyber, Argon represents a premium tier. Specialized or high-volume classifications that do not require complex reasoning remain significantly cheaper on smaller models, making a multi-tier routing architecture the most cost-effective production pattern.

Worked Monthly Budget for an Enterprise Pipeline

To evaluate Gemini 4 Argon pricing in an operational setting, consider an enterprise software engineering team running continuous code audits, pull request reviews, and security vulnerability patching. The calculation illustrates the financial impact of active context caching over a standard 30-day billing cycle.

Assume the organization processes 200 complex codebase tasks each business day, totaling 4,400 runs per month. Each run uses a 200,000-token repository context window, of which 180,000 tokens are cached core libraries and 20,000 tokens are fresh diffs, resulting in an average output generation of 10,000 tokens per task.

Without context caching, the 4,400 runs would consume 880 million standard input tokens ($1,760) and 44 million output tokens ($440), totaling $2,200 per month. With the 95% caching discount applied to the 792 million cached tokens ($79.20) plus 88 million fresh tokens ($176.00), total monthly input spend falls to $255.20, bringing the full monthly compute bill to $695.20.

• Monthly runs: 4,400 codebase audit and transformation tasks.
• Input volume: 88 million fresh tokens ($176.00) plus 792 million cached tokens ($79.20).
• Output volume: 44 million generated reasoning tokens ($440.00).
• Total net API expenditure: $695.20 per month with caching active, down from $2,200 without caching.

Standard Rates After the Introductory Period

A footnote in Google's official announcement lists standard rates of $4 per million input tokens and $20 per million output tokens, double the introductory rates. The introductory window runs through November 30, 2026, and the standard rates, including $0.20 per million cached input tokens, take effect on December 1, 2026.

Applying the list rates to the monthly workload of 4,400 runs doubles the baseline infrastructure spend. Without context caching, processing 880 million input tokens at $4 per million costs $3,520.00. Generating 44 million output tokens at $20 per million adds $880.00, bringing the un-cached total to $4,400.00 per month, exactly double the $2,200.00 introductory figure. From December 1, cached inputs cost $0.20 per million tokens. At that rate, 792 million cached tokens cost $158.40 and 88 million fresh tokens cost $352.00, for a monthly total of $1,390.40 including output fees.

• Standard input rate after the introductory window: $4 per million tokens.
• Standard output rate after the introductory window: $20 per million tokens.
• Un-cached monthly workload expenditure: $4,400.00 across 4,400 tasks, up from $2,200.00.
• Monthly workload expenditure with caching from December 1: $1,390.40.

Who This Pricing Suits and When to Route to Smaller Models

Gemini 4 Argon pricing is justifiable for high-stakes, multi-step workflows where reasoning failures carry direct operational risk. Tasks such as migrating large C/C++ codebases to Rust, conducting deep financial portfolio analysis, or patching zero-day security exposures benefit from the model's 1,048,576-token combined input and output window and verified benchmark performance.

This rate card is not suitable for routine, low-complexity interactions. Organizations looking to run basic conversational chatbots, simple text summarization, metadata extraction, or high-frequency customer ticket tagging should not route traffic to Gemini 4 Argon.

Across the workflows we have automated for SMB teams, the most common integration defect is sending routine queries to a high-cost frontier model. Teams that deploy a routing gateway to filter simple classification requests to smaller utility models protect their budget while reserving Gemini 4 Argon for complex multi-step reasoning.

Implementation Conditions: What Would Change This Assessment

Our pricing recommendation depends on the December 1, 2026 rate change and the availability of context caching. When standard pricing takes effect on December 1, 2026, doubling rates to $4 input and $20 output per million tokens, organizations running long-context generation loops will need to recalibrate their unit economics.

Similarly, if an enterprise application cannot structure requests to benefit from context caching, the higher baseline token rates will double or triple monthly operating expenses. Organizations must test whether their prompts share sufficient context to capture the 95% caching discount before committing to large-scale migrations.

Before deploying automated agents across your production systems, calculate your expected output token ceiling and audit your API spend against the official Google DeepMind rate card.