How to call Gemini 4 Argon with the Interactions API

The model id is already gemini-4-argon. Use the Gemini API guide as the call reference, not as a launch note. Last updated 2026-10-16 UTC, the page marks the Interactions API as the recommended path. generateContent is a separate tab. This article stays on Interactions, and it is not the Gemini 3.8 Flash guide.

The page calls Gemini 4 Argon its latest, most capable model, available through the Gemini API and in Google AI Studio. It is for longer reasoning on hard coding, agentic, math, and science tasks. Input, thinking, and output share one window of 1,048,576 tokens. Thinking levels are low, medium, and high.

Make the first Interactions call

The quickstart sends one text input to gemini-4-argon. The note above the samples says to use the recommended default thinking level, medium, and a 64K output token limit. The samples do not set either field. Python prints the interaction output text. JavaScript logs the same field. The curl sample does not show a read step. Leave both fields unset if you want those documented defaults. For standard tasks, the page later says to keep 65,536 or to set no output limit. It states both figures and does not reconcile them.

The page does not show a package install or how to create an API key. The Python sample constructs a client with no arguments. The JavaScript sample constructs a client with an empty object. The REST sample sends the key in a header and reads it from the environment variable written in that example. Use the key setup you already have.

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)

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

const client = new GoogleGenAI({});

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

console.log(interaction.output_text);

curl "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H 'Content-Type: application/json' \
  -X POST \
  -d '{
    "model": "gemini-4-argon",
    "input": "Write a three.js script that renders a realistic 3D black hole with gravitational lensing."
  }'

Replace the input string with your task. Keep the model id as gemini-4-argon.

Argon is not on the Free tier. Introductory pricing runs through November 30, 2026: $2.00 per 1 million input tokens and $10.00 per 1 million output tokens, including thinking tokens. Starting December 1, 2026, those rates are $4.00 and $20.00. Implicit caching is on by default for prompts of at least 4,096 tokens and discounts cached input by 95%, to $0.10 per 1 million through that window and $0.20 after it.

Set a thinking level

thinking_level controls quality, latency, and token use. max_output_tokens is a hard cap on thinking and output together. The response stops when that combined total is hit, so a tight cap can cut the answer short. To use fewer tokens, lower the thinking level.

• low uses the fewest tokens and the lowest latency. It fits simpler tasks, cost-sensitive work, and light analysis or writing. Quality can drop on hard tasks.
• medium is the default and the recommended level for most tasks, including complex code and agentic use.
• high uses the most tokens and has the slowest first token. Use it for complex reasoning, hard math and science, and the hardest code or agent tasks. It pairs with long decode for long outputs.

Test all three, then keep the lowest level that meets your quality bar.

This call sets medium in the generation config object. The comment is part of the published Python sample. Copy the JavaScript and REST blocks on the same page whole. They set thinking_level the same way.

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)

Output-cap starting points from the page:

• Standard tasks: keep 65,536, or do not set a value.
• A limit already below 32K: raise max_output_tokens to at least 32,768. Argon can spend more of that limit on thinking than earlier models.
• Complex reasoning and coding: turn background execution on and set max_output_tokens to 131,072 or 262,144, up to 1,048,576 minus the input.

Long decode can pass 64K tokens, up to 1 million minus the input, when background is true and max_output_tokens is greater than 65,536. Those replies are written in hops of up to 64K tokens. Each hop adds a model_output step. Join that text when the interaction completes. Copy the guide's polling sample, which uses high and a 262,144 cap. Do not shorten the loop.

Point Antigravity at the model

The Antigravity agent in Gemini Managed Agents supports Gemini 4 Argon. Set the model in agent_config. This sample does not use a top-level model field. It sets the agent id shown below, environment to remote, and an agent config whose type is antigravity and whose model is gemini-4-argon.

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    agent="antigravity-preview-09-2026",
    input=(
        "Audit https://web.dev for performance, Core Web Vitals, and SEO. "
        "Query Google's PageSpeed Insights API for both Mobile and Desktop strategies. "
        "Check search indexing with Google Search for site:web.dev. "
        "Format the output as a side-by-side scorecard table with prioritized fixes."
    ),
    environment="remote",
    agent_config={
        "type": "antigravity",
        "model": "gemini-4-argon",
    },
)

print(interaction.output_text)

The JavaScript sample sends the same fields and adds a second argument that sets timeout to 300000. The page does not state the unit, so leave the number as written. Hard prompts can take several minutes. Use background execution, or set HTTP client timeouts to at least 15 minutes. The page does not say that 300000 is that 15-minute setting. Copy the REST sample for the same agent, environment, and agent config fields. The linked model-selection section was not in this fetch, so no extra agent fields are added here.

Move an existing request

• Change the model string to gemini-4-argon.
• If max_output_tokens is set, consider raising it to at least 32,768.
• Choose low, medium, or high, and test all three.
• Drop explicit caching. cachedContents and caches.create are not supported. Do not create or reference cached_content.
• Remove temperature, top_p, and top_k. If sent, they are ignored.
• For output past 64K tokens, set background to true and max_output_tokens greater than 65,536.
• Use previous_interaction_id for multi-turn and function-calling conversations. The page has no sample for that field.

What to try next

Copy the Interactions sample for your language from the Gemini 4 Argon guide and run it with your own input before you set a thinking level. Then run one real prompt at low, medium, and high, and keep the lowest level that holds quality. For the managed agent, copy the Antigravity sample, including the remote environment and agent_config, instead of only changing the model string. If you need more than 64K tokens, copy the background-execution example on that page and join every model_output step.

Sources

• What's new in Gemini 4 Argon