Gemini 4 Argon: Fairwind, API Access and What It Costs

1. Two doors: Fairwind and the public API

The September 30 post said Argon would reach trusted cyber defenders first, then paid API customers and Google AI Ultra subscribers “as soon as possible”, with no model ID and no date. That was accurate that day. It stopped being accurate on October 13: the Gemini API changelog records gemini-4-argon as generally available, and Google’s second post, titled “available for everyone”, lists the paid tier in AI Studio, Google Cloud’s Gemini platform, Antigravity for AI Ultra, and the Gemini app and AI Mode in Search.

Fairwind did not go away. It is a separate door: the same model with cyber guardrails removed, for vetted defenders. According to Google Cloud’s announcement, the public model ships with cyber mitigations that block offensive tasks such as pentesting and exploitation. For most builders, the public API is the door that matters.

Who gets what (as of October 14, 2026)

2. What actually changed

The headline is output: from a 64K ceiling to 1M tokens. Read the docs before you celebrate. The default max output is still 65,536, and the 1M ceiling needs background execution (“long decode”), where the server chains hops of up to 64K tokens each. The window is also unified: input, thinking and output share the same 1,048,576 tokens.

The second change is positioning. At $2/$10, Argon costs about 2.7x Gemini 3.8 Flash per token ($0.75/$3.75 intro), and the ratio holds after both promos end ($4/$20 against $1.50/$7.50). You are paying for capability on hard tasks, not for a cheaper token.

Three changes that matter

3. Evidence, with sources

Every number below comes from Google’s model card for Argon (results as of October 2026) and is vendor-reported: Google chose the comparison set and the card gives no evaluation settings. It compares Argon with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5.

Selected rows, Argon vs the field

4. Quickstart and honest limits

The code below is adapted from Google’s Gemini API docs (what’s new in Gemini 4 Argon, updated October 14, 2026); I have not executed it. It uses the Interactions API, which is what the current Gemini quickstart uses. You need a GEMINI_API_KEY from a project with the paid tier.

Two things differ from older Gemini code. thinking_level goes inside generation_config, and anything above the 65,536 default needs background mode plus an explicit max_output_tokens. Argon also runs as the model behind the Antigravity managed agent (a preview agent); the docs cover it and I left it out here.

# pip install -U google-genai
# export GEMINI_API_KEY="YOUR_API_KEY"
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Review this retry logic for race conditions and propose a safer lock.",
    generation_config={
        "thinking_level": "medium"  # low | medium (default) | high
    },
)

print(interaction.output_text)

// npm install @google/genai
import { GoogleGenAI } from "@google/genai";

const client = new GoogleGenAI({});

const interaction = await client.interactions.create({
  model: "gemini-4-argon",
  input: "Review this retry logic for race conditions and propose a safer lock.",
  generation_config: {
    thinking_level: "medium", // low | medium (default) | high
  },
});

console.log(interaction.output_text);

import time
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Write a complete, rigorous proof of the prime number theorem.",
    background=True,
    generation_config={
        "max_output_tokens": 262144,
        "thinking_level": "high",
    },
)

while interaction.status in ("in_progress", "queued"):
    time.sleep(5)
    interaction = client.interactions.get(id=interaction.id)

# Each hop adds a model_output step, so join the text of all of them.
text = "".join(
    content.text
    for step in interaction.steps
    if step.type == "model_output"
    for content in step.content
    if content.type == "text"
)
print(text)
print("Status:", interaction.status)

5. Pick your door this week

Start from who you are, not from the benchmark table. Most readers belong on the public API. Only a narrow group qualifies for Fairwind, and Google publishes no approval timeline.

Verdict

Argon is a real step up for hard, long-horizon engineering and long-context work, and it raises the output ceiling from 64K to 1M tokens. It is also a premium model with a promo price, verbose reasoning and a list of unsupported knobs. The benchmark picture is mixed, not a sweep.

My call: put it on the slice of tasks where Flash-class models fail, keep Gemini 3.8 Flash for volume, and re-price everything at $4/$20 before December. If you are a defender, apply to Fairwind now. If you are not, the public API is the door that is open today.