What GitHub announced
GitHub describes Argon as built for complex coding, agentic and reasoning tasks. It says early testing showed the model solving more tasks than earlier Gemini models on every coding benchmark, with the biggest gains in Copilot CLI. It also says the model checks its own work and tests edge cases before it finishes.
Those are GitHub's claims. The changelog doesn't name the benchmarks, the task counts, the comparison models or the size of the gains, so treat them as vendor assertions until someone reproduces them.
Independent tracking exists but is thin. One roundup reports that Artificial Analysis scores Argon 53 on its Intelligence Index, tied with GPT-6 Astra and one point above GPT-6.1 Sol. The same roundup says Argon uses about 62K output tokens per index task against 27K for Astra. That detail matters for cost, as covered below. These are Argon's general-purpose scores, not Copilot-specific results.
Who gets it, and where
Argon is available to Copilot Pro+, Max, Business and Enterprise users. GitHub does not list Copilot Free or Pro.
You can pick it in the model picker in these clients:
- Visual Studio Code
- Visual Studio
- Copilot CLI
- GitHub Copilot coding agent
- The GitHub Copilot app
- JetBrains IDEs
- Xcode
- Eclipse
Rollout is gradual, so an eligible account may not see it right away. If it's missing, check back later. Changing settings won't speed up the rollout.
The announcement says nothing about making Argon a default model. It also gives no minimum client versions. Argon is an option you choose, not a replacement for whatever Copilot uses by default.
Admin controls for Business and Enterprise
Copilot Business and Enterprise administrators manage access through the model policy in Copilot settings. New models are enabled automatically under default model enablement. That stops only if an administrator has turned off the global default or explicitly disables this model.
Organizations that want to review models before developers use them should check two things:
- The global default. If it's on, Argon is enabled automatically for eligible users as it rolls out.
- The Argon-specific policy. Disable it explicitly if you want a hold regardless of the default.
GitHub gives no click-by-click steps in the announcement. Look in the Copilot policy settings for your organization or enterprise.
Billing: the part to read carefully
GitHub says Argon is billed at provider list pricing under usage-based billing. GitHub's pricing documentation explains the mechanism. Each interaction consumes input, output and cached tokens, and the cost is converted to AI credits at 1 credit = US$0.01. Individual plans include AI credit allowances. Business and Enterprise have per-user allowances pooled at the billing-entity level. Usage beyond the allowance is billed per token.
The GitHub pricing table lists Gemini 4 Argon as GA, in the "Powerful" category, with a default tier and no long-context tier. The listed rates per million tokens are:
| Token type | Rate per 1M tokens |
|---|---|
| Input | $2.00 |
| Cached input | $0.10 |
| Output | $10.00 |
A footnote says these are promotional rates that run until December 1, 2026. GitHub's page doesn't say what the rate becomes afterward. Other coverage of Google's own API pricing says Argon costs $2 and $10 per million input and output tokens until November 30, 2026, and the price doubles to $4 and $20 from December 1. That is Google's API schedule. Don't assume Copilot will follow it, but budget for a possible increase.
Output dominates the bill in agentic work. If Argon really uses many output tokens per task, as the Artificial Analysis figure above suggests, the effective per-task cost can differ from what the per-token rate implies. Code completions and next edit suggestions aren't billed in AI credits, according to GitHub's docs, so the cost applies to chat, agent and CLI-style usage.
Google's API specs are not Copilot guarantees
Research notes cite Google documentation describing a context window of about 1 million tokens and multimodal inputs. I couldn't verify those pages directly, and GitHub doesn't say Copilot exposes them. Don't assume a million-token context, image or video input, or Google's API thinking-level settings are available inside Copilot. Test the model in your own client.
Practical advice
- Individuals on Pro+ or Max: open the model picker in your preferred client and look for Argon. If it isn't there, wait.
- Teams: try it on a bounded task, such as a test-heavy refactor in Copilot CLI, and compare credit consumption against your current model before standardizing.
- Admins: decide on the global default and the Argon policy before the rollout reaches your users. Tell finance that the promotional rate has an end date.
- Skeptics: GitHub's "solved more tasks" claim compares Argon with earlier Gemini models, not with competing vendors' models. It doesn't show Argon beating your current model.
Bottom line
Argon is a legitimate new option for paid Copilot tiers across the major IDEs and the CLI. The performance case rests on GitHub's own unspecified testing. Admins, though, get clear controls, and the pricing has a visible promotional window that ends December 1.