Gemini 4 Argon Lands in Cursor and Vercel's AI Gateway, Matching Claude Opus 5.5 on CursorBench

Google made Gemini 4 Argon generally available on October 13, 2026, rolling it out across the Gemini API, Google Antigravity, Gemini for business and the Gemini app. That same day, Cursor added the model to its picker — confirmed by a post from Cursor’s own account on X timestamped 7:17 PM on October 13. Vercel’s AI Gateway also added Argon as a routable provider reachable from Claude Code, Codex and Cursor; Vercel’s changelog entry does not carry its own publish date in our capture, but it confirms the model is live there now, within the same short window as the Cursor and Google rollouts.

For anyone deciding which model to point a coding agent at this week, that quick arrival across tooling is the practical story. Argon is now one model name away inside tools many teams already run, at a published price and a benchmark result that puts it in the same tier as Claude Opus 5.5.

What Cursor shipped

Cursor’s model docs list Gemini 4 Argon with a 200k working context window (expandable to a 1M-token max), four effort levels — high, medium (the default), low, and minimal — and access to Cursor’s full agent toolset: file search and edits, shell commands, browser control, and image generation. It is priced at $2 per million input tokens and $10 per million output tokens, with cached input reads discounted to $0.10 per million.

Cursor’s docs note that plans include two usage pools, and Gemini 4 Argon draws from the third-party Other Models pool, metered at the per-token rates listed above.

On capability, Cursor said on X that Argon scores 52.8% on CursorBench at default effort — “on par with Opus 5.5” — at that same $2/$10 per-million-token price. CursorBench is Cursor’s own internal benchmark, not an independently audited cross-lab suite, so the tie is best read as a vendor-reported first filter rather than a settled ranking. It is still a useful signal when triaging which model to assign to a given coding lane, especially paired with the four effort levels Cursor exposes for the same model.

What Vercel shipped

Vercel’s AI Gateway changelog confirms Argon is live there as google/gemini-4-argon, callable through the AI SDK, the Chat Completions, Responses, and Anthropic Messages APIs, or any coding agent wired to the gateway. Unlike Cursor’s effort levels, Vercel exposes Argon’s reasoning control as thinking levels — low, medium, and high — set via reasoning in the AI SDK or reasoning_effort in Chat Completions, rather than a numeric thinking-token budget.

To use Argon from an existing coding agent rather than writing raw API calls, Vercel’s instructions are to install the latest CLI and run vercel ai-gateway setup, then select google/gemini-4-argon inside Claude Code, Codex, or Cursor. The gateway reflects Google’s own pricing with no added markup and no platform fee, including on bring-your-own-key requests, which matters for teams already paying Google directly and just want Argon inside a unified usage and cost dashboard alongside their other gateway models.

Why the price window matters

Argon’s $2/$10 per-million-token rate is introductory. Google’s own announcement states the price doubles to $4 input / $20 output after November 30, 2026, for the publicly accessible version of the model — a schedule worth budgeting now rather than discovering on the invoice (our price-expiry register walks through how to track that kind of two-tier pricing). Full details of Google’s own rollout, including Antigravity and enterprise access, are in our earlier coverage of the Argon launch.

What to actually do with this

If you’re already paying for Cursor and want a cheaper frontier-tier option alongside Opus 5.5, Argon is now a model-picker selection away, with effort levels to trade speed for depth on a per-task basis. If your agents already route through Vercel’s AI Gateway for Claude Code or Codex, adding Argon is a vercel ai-gateway setup run and a model-name swap rather than a new integration — useful if you’re already comparing routing strategies the way we described in Routers You Can’t See Inside. Either way, treat the CursorBench tie with Opus 5.5 as a reason to try Argon on a real coding lane before standardizing on it, not as a settled verdict — and revisit the choice before the November 30 price change lands.

Load X post

Loading this post connects to X.

The original source is available above.