Gemini 4 Argon Guide: API Pricing (Intro Rate Through Nov 30), Specs, and Migrating from Gemini 3

What Gemini 4 Argon Is: The New Flagship, GA on October 13

According to Google's official changelog, Gemini 4 Argon became generally available on October 13, 2026, described as "our most capable model built for complex coding, agentic, math, and science tasks." The model page is more specific: it is built for complex software engineering, deep mathematical and scientific problem solving, autonomous multi-step agents, and long decode up to 1M tokens.

It is currently the only model in the Gemini 4 family. The Gemini 3 family remains in service, and 3.8 Flash (GA on 2026-09-02) and 3.5 Flash-Lite are the models Google recommends for new projects.

Current Gemini Text Models and Pricing

Source: official Gemini API models, pricing, and changelog pages (checked 2026-10-28), USD, Standard consumption. Gemini 3.7 Flash and 3.5 Flash were deprecated on 2026-10-08 and are not listed.

Pricing: Introductory Through 11/30, Double from 12/01

All four public consumption options for Argon have two price stages on the same dates:

Things that are easy to miss:

• Output price includes thinking tokens: tokens the model spends on internal reasoning are billed at the output rate
• Hidden grounding tokens are billed too: Google states that internal tool system instructions and function declarations are billed when you use grounding
• Google Search grounding is billed separately: 5,000 free search requests per month (shared across all Gemini 3 and newer models), then $14 per 1,000 requests
• No free tier: every Free Tier cell reads "Not available"; content on the paid tier is not used to improve Google's products, except on the invite-only Contributor tier (a private preview), where content is used to improve products

Key Specs

• Context: 1,048,576 tokens, shared by input, thinking, and output
• Output per response: up to 65,536 in default synchronous mode; with background execution and max_output_tokens above 65,536, generation can run up to "1M minus your input"
• Inputs/outputs: text, image, video, audio, and PDF in; text out
• Supported: function calling, structured outputs, code execution, Google Search and Google Maps grounding, URL context, file search, computer use (Preview), agentic video understanding (defaults to static)
• Not supported: image generation, audio generation, Live API
• Consumption options: Batch, Flex, and Priority are all supported
• Managed Agents: the Antigravity agent in Gemini Managed Agents can be set to use Argon

Thinking Levels: The Main Dial for Quality, Latency, and Tokens

max_output_tokens is a hard limit on thinking plus output combined: the response stops when it is reached. For existing apps with a limit below 32K, Google recommends raising it to at least 32,768; to use fewer tokens, lower the thinking level rather than the limit.

Quickstart: The Official Interactions API Example

This is Google's minimal example (Python, using the default medium level and a 64K output limit):

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Write a three.js script that renders a realistic 3D black hole with gravitational lensing."
)

print(interaction.output_text)

To set the thinking level, put it in generation_config:

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Analyze this payment processing pipeline for race conditions during retry attempts and rewrite the transaction locks safely.",
    generation_config={
        "thinking_level": "medium"  # Balanced reasoning effort for complex tasks
    }
)

print(interaction.output_text)

For outputs longer than 64K, add background=True and set max_output_tokens (for example 262144). The model generates in hops of up to 64K tokens, each hop adds a model_output step to interaction.steps, and you join the text across all of them when the interaction completes. If you set background without max_output_tokens, the server stops after 64K.

Cost Math (Arithmetic, Not Benchmarks)

Assume 50 million input tokens and 10 million output tokens (including thinking) per month, no caching, Standard consumption:

If 40 million of the 50 million input tokens hit the implicit cache, Argon's input cost during the introductory period is $4 (cached) + $20 (uncached) = $24, for a monthly total of $124; from 12-01 it is $8 + $40 = $48, for a monthly total of $248. Implicit caching only applies to prompts of at least 4,096 tokens, and hit rates depend on your actual requests. The same volume on Batch costs $100 a month on Argon during the introductory period and $200 from 12-01.

⚠️ Two caveats. First, December is the month with the widest gap between the two: Argon has already doubled while 3.8 Flash is still on its introductory price, so the unit price gap reaches 5.3x (about 2.7x in other periods). Second, Google notes that Argon can spend more of the output limit on thinking than earlier models, so the same task may produce more output tokens on Argon. Don't compare unit prices alone.

Migrating from the Gemini 3 Family: The Official Checklist

Google's migration checklist lists the items below; the ones near the top are the most likely to require changes to existing code:

1. Update the model ID: change your target model string to gemini-4-argon
2. Move off explicit caching: Argon does not support cachedContents / caches.create, so remove calls that create or reference cached_content; implicit caching is on by default (prompts of at least 4,096 tokens), and cache hits cost 5% of the regular input price
3. Review max_output_tokens: if you set it explicitly, Google recommends at least 32,768
4. Configure the thinking level: test low, medium (default), and high before deciding
5. Remove sampling parameters: temperature, top_p, and top_k are ignored if provided
6. Use background execution for long outputs: for tasks over 64K, set background to true and max_output_tokens above 65,536
7. Increase timeouts: hard prompts can take several minutes; use background execution or set HTTP client timeouts to at least 15 minutes
8. Keep conversation state on the server: use previous_interaction_id for multi-turn and function calling conversations
9. Review cost estimates: every token the model processes is billed, including internal tool instructions when using grounding

Two related changes from the same period:

• 3.7 Flash and 3.5 Flash are deprecated (2026-10-08): they can no longer be selected in AI Studio, and API requests are automatically routed to gemini-3.8-flash / gemini-3.6-flash; the endpoints themselves are not shutting down. Google recommends updating your model string to the newer version
• Request storage is on by default: the Interactions API stores requests by default; since 2026-10-26, new paid projects using the Generate Content API also default to store=true, which you can turn off per request with store=false or in the Logs and Datasets settings in AI Studio

CloudInsight Enterprise Gemini API Procurement | Local Invoices, No Foreign Credit Card Needed

With the introductory price ending in late November, what enterprises get stuck on most is often not the technology but payment, invoicing, and budgeting across billing months.

Contact CloudInsight for a Gemini API Enterprise Quote

We offer one-stop procurement for Claude, GPT, and Gemini, Taiwan uniform invoices, and technical support in Chinese.

How to Choose: Splitting Work Between Argon and 3.8 Flash

A common pattern is tiering: route everyday traffic to 3.8 Flash, escalate only the problems Flash handles poorly to Argon, and default to low/medium, turning on high only for a small set of tasks.

FAQ: Gemini 4 Argon

How much does Gemini 4 Argon cost?

$2 per million input tokens and $10 per million output tokens (including thinking), with cached input at $0.10, through 2026-11-30; from 2026-12-01 it is $4/$20, with cached input at $0.20. Batch and Flex are half of Standard.

Does Gemini 4 Argon have a free tier?

No. Argon is not available on the free tier, so billing starts with the first request. For free testing, start with Gemini 3.8 Flash, which has a free tier.

What happens to pricing after November 30?

From 2026-12-01, Standard is $4 input, $20 output, and $0.20 cached input; Batch and Flex are $2/$10; Priority is $7.20/$36. The same usage roughly doubles your bill.

Gemini 4 Argon or Gemini 3.8 Flash?

Choose Argon for the deepest reasoning, long outputs, or the hardest agent tasks; start general agent and coding work on 3.8 Flash. 3.8 Flash also has introductory pricing ($0.75/$3.75 through 2026-12-31, $1.50/$7.50 from 2027-01-01) and a free tier.

Does Gemini 4 Argon support context caching?

Implicit caching only. It is on by default for prompts of at least 4,096 tokens, and cache hits cost 5% of the regular input price. Explicit caching (cachedContents / caches.create) is not supported.

Can we still use Gemini 3.7 Flash and 3.5 Flash?

Both were deprecated on 2026-10-08 and can no longer be selected in AI Studio. API requests are automatically routed to 3.8 Flash and 3.6 Flash, but Google recommends updating the model string directly.

What code changes does Argon need?

The three most common: update the model ID, remove explicit caching calls, and raise max_output_tokens to at least 32,768. Sampling parameters (temperature and others) are ignored, so you can remove them too.

Conclusion: Test During the Intro Period, Tier Your Usage from December

Three Core Recommendations

1. Finish evaluations before the end of November: run a low/medium/high comparison while the introductory price lasts, at half of what it costs from December
2. Budget at the 12-01 price: $4/$20 is the long-term price; treat the introductory price as a testing window
3. Tier your usage: route everyday traffic to 3.8 Flash and give only the hardest work to Argon; remove explicit caching before migrating

Ready to Adopt Gemini 4 Argon?

Contact the CloudInsight sales team and we'll help plan your Argon and 3.8 Flash usage mix and handle payment and invoicing.

Join our LINE Official Account for instant model-selection consultation.

---

Further Reading

• Gemini API Pricing Guide
• What Is Gemini API? Complete Guide
• Gemini API Python Tutorial
• LLM Model Ranking & Comparison
• How to Buy AI API in Taiwan

References

• What's new in Gemini 4 Argon — Gemini API Docs (checked 2026-10-28)
• Gemini 4 Argon model page — Gemini API Docs (checked 2026-10-28)
• Gemini Developer API pricing (checked 2026-10-28)
• Gemini API models
• Gemini API release notes (changelog)
• Gemini API deprecations