Skip to content
NewsProduct launch

GitHub adds Gemini 4 Argon to Copilot's model picker, two weeks after Google called it pilot-only

· by Pondero Newsdesk

The short version

GitHub's October 13 changelog made Google's Gemini 4 Argon generally available across eight Copilot surfaces for paid plans, days after Google said broader API access would wait on a government review with no fixed date.

GitHub adds Gemini 4 Argon to Copilot's model picker, two weeks after Google called it pilot-only

GitHub made Google's newest frontier model, Gemini 4 Argon, generally available across eight Copilot surfaces on October 13, with no preview tag and no waitlist, per GitHub's changelog. Google introduced Argon on September 30 by restricting it to a vetted security pilot and declining to set a date for wider release, according to Google's own announcement.

What happened

Gemini 4 Argon now appears in the model picker inside Visual Studio Code, Visual Studio, Copilot CLI, the Copilot coding agent, the Copilot app, JetBrains IDEs, Xcode, and Eclipse, GitHub said. Access is limited to the Copilot Pro+, Max, Business, and Enterprise plans; the entry-level Copilot Pro tier is not on the list. Usage bills at Google's own list pricing under Copilot's usage-based billing system rather than a flat per-seat rate, and GitHub is rolling the model out gradually, telling customers who do not see it yet to check back later. For Business and Enterprise accounts, admins control access through Copilot's model policy settings, and GitHub's default behavior enables a new model automatically unless an admin has turned off global default enablement or blocked Argon by name.

Argon's Copilot debut skips the gate Google built for the model itself

Google did not frame Argon's September 30 release as a broad rollout. The company said Argon's standout capability, autonomously finding and patching critical software vulnerabilities, would go first to a small group of vetted defenders in its Fairwind Program, and that public API and Google AI Ultra access would wait on a voluntary US government pre-release review with no completion date attached, per Google's announcement. Copilot's integration is a separate product built for general coding and agentic tasks rather than Fairwind's vulnerability-hunting workflow, and GitHub's changelog makes no claim that it carries that workflow over. But the underlying model behind both products is the same release, and GitHub shipped broad commercial access to it two weeks after Google described its own path to broad access as still pending review. Admins who rely on Copilot's auto-enable default should decide now whether a model with an unresolved security review elsewhere belongs in their organization's policy, rather than discovering it already turned on once the gradual rollout reaches their account.

The billing mechanics carry a practical cost wrinkle too. Google's announcement set Argon's price at an introductory $2 per million input tokens and $10 per million output tokens, stepping up to $4 and $20 once that introductory window closes, per a footnote in Google's post. Because Copilot bills Argon usage at provider list pricing, any team that adopts it through usage-based billing inherits that same step-up whenever Google ends the discount, with no separate notice from GitHub required.

The model itself also changed in a way that matters for agentic coding specifically. Google raised Argon's maximum output to 1 million tokens, up from 64,000 tokens on the prior Gemini generation, a 16x jump the company says lets the model work through one long, multi-step task without breaking it into separate calls, per Google's announcement. That change applies to the same model GitHub just added to Copilot, so a long-running Copilot coding agent session, not just Fairwind's security workflow, is the kind of task the larger output window was built for.

Context and reactions

Argon arrived with benchmark claims that have not been independently reproduced. Google cited a 77.9% score on the DeepSWE v1.1 long-horizon coding test and a tie for first place at 68% on CWE-bench v1, a vulnerability-patching benchmark, according to Google's release. GitHub's characterization of the Copilot version is similarly self-sourced: the company says that in early testing, Argon solved more tasks than prior Gemini models on every coding benchmark it checked, with the largest gains showing up in Copilot CLI, and that the model tests its own edge cases before finishing a task, per GitHub's changelog. Neither claim comes with a published scorecard a Copilot admin can check directly.

The addition also lands in a month where Copilot's model lineup has moved repeatedly. GitHub swapped the replacement model for a retiring Claude Opus 4.7 twice inside a month, first naming Claude Opus 5 on September 3 and then shipping Claude Opus 5.5 in its place on October 2, per GitHub's changelog. Argon's arrival is the fourth distinct model-lineup change GitHub has pushed to Copilot admins in under two weeks.

Outside Copilot, the one disclosed result from Argon's restricted access so far involves a different kind of partner. Security vendor Wiz, which has Fairwind access through a free public-infrastructure audit program, used Argon to find a vulnerability exposing patient data across hospital software used worldwide, a flaw Google says earlier frontier models had missed, according to Google's announcement. That result came from the gated security workflow, not the general coding capability now sitting in Copilot's picker, but it is the clearest evidence so far that Argon's underlying model finds things prior versions did not.

What to watch next

Watch whether Google widens the Fairwind Program or opens general API and Ultra access before Argon's introductory pricing window closes, since Google tied the price jump to the end of that window rather than to a calendar date. Also watch Copilot's model-policy pages for whether Argon's rollout finishes evenly across all eight listed surfaces or lags behind on some of them, since GitHub has committed only to a gradual rollout without a surface-by-surface timeline.

Sources