Gemini 4 Argon API: Pricing, Context Limits and Code Examples

Gemini 4 Argon is Google's new flagship model, announced on September 30, 2026 and opened to developers on October 13, 2026. On Onysoft AI Gateway it is available as google/gemini-4-argon: as of October 17, 2026 the list price is $3 per 1M input tokens (≈148.26 TL) and $15 per 1M output tokens (≈741.31 TL), with a 1,048,576-token context window. You can call it with your existing OpenAI SDK code, paying from a Turkish Lira balance — no foreign card or separate cloud account required.

This guide brings together the facts about Argon verified against official sources (dates, input types, context and output limits), the USD and TL price table on Onysoft, what the end of the introductory pricing on November 30, 2026 means for your budget, and working Python/curl examples. Benchmark scores are quoted only as Google published them, with the source stated.

What Is Gemini 4 Argon? Dates and Technical Facts

Google presents Argon as a frontier model built for real-world software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense, focused on sustained reasoning across long, multi-step tasks. The model at a glance:

Item	Detail	Source
Announced	September 30, 2026	Google DeepMind announcement
First access	Trusted cyber defenders in the Fairwind Program	Google DeepMind announcement
Opened to developers	October 13, 2026	Google developer announcement
Google model code / status	gemini-4-argon / Stable	Gemini API model page
Onysoft ID	google/gemini-4-argon	Onysoft catalog
Input / output	Text, image, video, audio, PDF / text	Gemini API model page
Context	1,048,576 tokens (one budget shared by input and output)	Gemini API model page, Onysoft catalog
Max output	65,536 tokens by default; up to 1,048,576 in long decode mode	Gemini API model page
Thinking levels	low, medium, high	Gemini API model page

The headline change is the output limit: Google says Argon raises the output token limit to 1 million, up from 64K. According to the model documentation that ceiling belongs to long decode mode; a standard request defaults to 65,536 output tokens, and the context budget is shared with the input, so a very long prompt leaves less room for the response. When you need long outputs through Onysoft, set max_tokens explicitly and verify the limit with a small test.

Google also described safety measures for Argon, including defenses against cyber and CBRN misuse and stronger resistance to indirect prompt injection. That matters if you plan to use Argon in agent flows that process external content, but it does not replace your own validation and permission checks.

Benchmark Results Published by Google

The scores below are Google's own claims from its September 30, 2026 announcement; Onysoft has not reproduced these measurements independently. Benchmarks measure specific task sets and are no substitute for performance on your own workload.

Benchmark	What it measures	Gemini 4 Argon (as reported by Google)
DeepSWE v1.1	Software engineering tasks	77.9% (a new state of the art, according to Google)
AutomationBench (Zapier)	Workflow automation	51.3% (ranked first as of September 30, 2026, according to Google)
LVBench	Long video understanding	91.7%
CWE-bench v1	Cybersecurity	68% (tied for first, according to Google)

Google also says Argon leads on the Vals Index, Vals Finance Agent v2 and Harvey's legal agent benchmark, but gives no numbers for those three. The fastest way to compare models on your own scenario is to run the same prompt side by side in the Playground.

Gemini 4 Argon Pricing: USD and TL Table

The table shows Argon and nearby models at Onysoft list prices as of October 17, 2026, per 1 million tokens:

Model	API ID	Context (tokens)	1M input	1M output	1M input (TL)	1M output (TL)
Gemini 4 Argon	google/gemini-4-argon	1,048,576	$3.00	$15.00	148.26 TL	741.31 TL
Gemini 3.1 Pro	google/gemini-3.1-pro-preview	1,048,576	$3.00	$18.00	148.26 TL	889.57 TL
Gemini 3.8 Flash	google/gemini-3.8-flash	1,048,576	$1.2375	$6.1875	61.16 TL	305.79 TL
Claude Opus 5.5	anthropic/claude-opus-5.5	1,000,000	$6.00	$30.00	296.52 TL	1,482.61 TL
GPT-6 Astra	openai/gpt-6-astra	1,050,000	$15.00	$75.00	741.31 TL	3,706.53 TL

Prices are Onysoft list prices as of October 17, 2026; see /models for current pricing. Lira figures use 1 USD = 49.4204 TL (TCMB rate of October 16, 2026); billing uses the TCMB rate at request time. On Gemini 3.1 Pro, prompts above 200,000 tokens are billed at a higher tier.

Monthly example: an app that processes 10 million input and 2 million output tokens a month pays, at today's list prices, $60 on Argon (≈2,965.22 TL), $66 on Gemini 3.1 Pro (≈3,261.75 TL, assuming prompts under 200,000 tokens), $24.75 on Gemini 3.8 Flash (≈1,223.15 TL) and $300 on GPT-6 Astra (≈14,826.12 TL). Use the cost calculator to run your own volumes.

Billing follows input and output token counts, with no subscription or commitment. Google counts thinking tokens as output, so requests that reason at length use more output tokens and cost more. Every response returns its cost in the cost field; track your budget from that field rather than from list prices.

Introductory Pricing Ends November 30: Budgeting for December

Google applies Argon's current pricing through November 30, 2026. According to the Gemini API pricing page, input, output, cached input and batch unit prices all double from December 1, 2026.

The Onysoft catalog updates prices through a daily sync, so Argon's Onysoft list price is expected to change by the same ratio around December 1, 2026: roughly $6 per 1M input tokens and $30 per 1M output tokens. The exact figure will appear on /models on that date.

• Budget: the monthly cost of the example app above rises from $60 today to about $120 (≈5,930.45 TL).
• Code: the price change does not change the ID; code written against google/gemini-4-argon keeps working.
• Cost levers: route speed-focused steps to a lower-priced model such as Gemini 3.8 Flash, and trim unnecessary context from prompts. See the AI API cost optimization guide for more techniques.

When to Use Argon, and When to Use the Others

The nature of the job and your budget decide. The mapping below is based on how Google positions Argon and on October 17, 2026 list prices:

Workload	Starting point	Why
Long-horizon agent tasks, code migrations, multi-step software engineering	google/gemini-4-argon	The area Google highlights for Argon; 1,048,576-token context.
Enterprise knowledge work such as legal and finance documents	google/gemini-4-argon	Google's positioning; long documents processed without chunking.
Speed-focused chat, real-time apps, high volume	google/gemini-3.8-flash	List price about 2.4 times lower than Argon.
Existing flows already tested on Gemini 3.1 Pro	google/gemini-3.1-pro-preview	Active in the catalog; migrate with side-by-side tests.

Argon or GPT-6 Astra? Both are flagship models with roughly 1-million-token context windows. The price gap is large: at October 17, 2026 list prices, Astra's token price is five times Argon's ($15 vs $3 per 1M input, $75 vs $15 per 1M output), and even after Argon's price doubles on December 1 the gap stays at 2.5 times. We do not make a general quality call; try both models with the same prompts in the Playground. For Astra details, see the GPT-6 Astra API guide.

First Request with Python and curl

To get started, create an account, generate an API key with the sk-ony- prefix in the dashboard and top up a lira balance. Because the Onysoft endpoint follows the OpenAI schema, the official openai Python package works directly in streaming mode:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.onysoft.com/v1",
    api_key="sk-ony-YOUR_KEY",
)

stream = client.chat.completions.create(
    model="google/gemini-4-argon",
    messages=[
        {"role": "user", "content": "Draft a step-by-step code migration plan for this service and note the risk of each step."}
    ],
    max_tokens=8192,
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Non-streaming requests: Onysoft returns non-streaming responses in a success/data envelope; the OpenAI-format response object sits in the data field. With the Python SDK you can read the raw body to see the content and token usage:

raw = client.chat.completions.with_raw_response.create(
    model="google/gemini-4-argon",
    messages=[{"role": "user", "content": "Summarize the termination and penalty clauses in this contract: ..."}],
    max_tokens=4096,
)
data = raw.http_response.json()["data"]
print(data["choices"][0]["message"]["content"])
print("input / output tokens:", data["usage"]["prompt_tokens"], data["usage"]["completion_tokens"])

The same request with curl. The content sits under the data field; for streaming, add "stream": true to the body:

curl https://api.onysoft.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ony-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-4-argon",
    "max_tokens": 4096,
    "messages": [{"role": "user", "content": "Review this error log and explain the likely root cause."}]
  }'

Image input and thinking depth: send images as OpenAI-style content parts (image_url). Google also documents video, audio and PDF input for Argon; run a small test with your own files before adding those types to a production pipeline. To control thinking depth, pass the reasoning parameter through extra_body in the Python SDK; it is forwarded unchanged to models that support it, so measure its effect from the token usage in the response.

See the API documentation for supported parameters and error codes, the model catalog for all 836 models available with the same key, and the Gemini models page for the whole family.