Gemini 4 Argon vs Claude Opus 5.5: which model for which job

Google says Gemini 4 Argon beats Claude at coding for half the price. That's half the story. Google's own model card has Argon ahead on DeepSWE and Opus 5.5 ahead on Terminal-bench, and Argon's half-price launch rate ends on 30 November 2026. This is the full version of the ARGON guide from my Reel. Facts checked on 20 Oct 2026 against ai.google.dev, deepmind.google and claude.com/pricing.

01What it is

Gemini 4 Argon (gemini-4-argon) is Google's new flagship reasoning model. The Gemini API docs list it as Stable and "now available", with a 1M-token context window, and describe it as built for complex software engineering, maths and science problems, and multi-step agents. Google published the model card on 13 October 2026.

02What the scores say (Google's numbers)

All of these come from Google's own model card. Google ran them, and we found no independent re-run yet.

DeepSWE v1.1 (fixing real code): Gemini 4 Argon 77.9%, Claude Opus 5.5 74.2%.
Terminal-bench 4.0 (agent work in a terminal): Claude Opus 5.5 66.4%, Gemini 4 Argon 57.4%.

So "beats Claude at coding" is only half true. Argon leads on one coding test, and Opus leads on the other by about nine points. The model card links to a methodology page for the harness and settings behind each score.

03What it costs (per 1M tokens, checked 20 Oct 2026)

Gemini 4 Argon, until 30 Nov 2026: $2 input / $10 output.
Gemini 4 Argon, from 1 Dec 2026: $4 input / $20 output.
Claude Opus 5.5: $4 input / $20 output.
Claude Sonnet 5.5: $2 input / $10 output.

Argon is half the price of Opus only until 30 November. From 1 December it costs the same as Opus. At its launch price it costs the same as Sonnet 5.5. Argon is not on the Gemini API free tier.

04Try Argon yourself

Get a Gemini API key at https://aistudio.google.com/apikey. You need a paid tier for Argon.
Set the key: export GEMINI_API_KEY="YOUR_API_KEY"
Install the SDK: pip install -U google-genai (or npm install @google/genai)
Call the model. This is the official quickstart call with the model name swapped to Argon:

Python quickstart call

from google import genai

client = genai.Client()
interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Find the bug in this function and fix it: ..."
)
print(interaction.output_text)

05Which model for which job

Fixing code in a repo on a budget: try Argon, especially before 30 November.
Long terminal and agent runs (multi-step shell work): stay on Opus 5.5.
After 1 December: the prices are the same, so pick on results alone.
Benchmarks are a starting point. Run both models on one of your own real tasks before you switch.

06Good to know

The scores are vendor-reported. Treat them as Google's claim until someone runs them independently.
Output price includes thinking tokens, so long reasoning answers cost more than the reply length suggests.