Gemini 4 Argon lands in Cursor at half the price of Opus 5.5

You can pick Gemini 4 Argon in Cursor's model menu now. Cursor announced it at 3:17 PM ET on Oct 13, saying it "scores 52.8% on CursorBench at default effort, on par with Opus 5.5." It is priced at $2 per million input tokens and $10 per million output tokens, which is half of Opus 5.5's $4 and $20 on Cursor's own price table.

What changed#

Google made Argon globally available at about noon ET. Logan Kilpatrick of Google said it is rolling out "in the Gemini API, to Pro + Ultra in Gemini App, and Ultra subs in Antigravity." Cursor added it about three hours later.

Cursor's model page lists these details:

• Model ID gemini-4-argon, from Google.
• A 200k context window by default and a 1M max context.
• Four effort levels: high, medium (the default), low and minimal.
• Access to all agent tools, including file search, edits, shell, browser and image generation.
• Pricing: $2 input, $0.10 cache read and $10 output per million tokens. No cache write price is listed.

Why it matters#

If you run Opus 5.5 for agent work, Cursor says Argon is level with it on Cursor's own benchmark at half the token price. For long agent runs, that could stretch your included usage further. The 52.8% figure is Cursor's number at default (medium) effort, and Cursor did not publish Opus 5.5's exact score in the post.

How to use it#

1. Update Cursor and pick Gemini 4 Argon from the model list. It starts at medium effort.
2. Set effort to high for hard multi-file work, or drop to low or minimal for quick edits, as the model page suggests.
3. Try it on a task you normally give Opus 5.5 and compare cost in your usage dashboard.

A prompt to run a quick side-by-side:

Using Gemini 4 Argon at medium effort, fix the failing test in this repo, explain the root cause in three sentences, and list every file you changed.

Good to know#

• Argon draws from the Other Models pool, not the Cursor Models pool, per the model page. The pricing page says the Start plan in India doesn't include that pool.
• You can't choose it in Grok Bot. Cursor's docs say "Grok Bot manages model selection, so there's no model picker."
• Cursor's post doesn't say whether contexts past 200k cost more. Check your usage before running 1M-token jobs.
• We have not tested Argon in Cursor from this box.

What to watch#

Whether independent CursorBench-style results hold up against Opus 5.5, whether Cursor publishes long-context pricing, and whether Argon shows up as an option in Auto or Cloud Agents.