For developers, the immediate questions are practical: Where does it appear, who controls access, and what happens to the usage bill?
Who gets Carbon—and where to find it
GitHub lists Copilot Pro+, Max, Business, and Enterprise as eligible plans. Users can choose Carbon through the model picker in:
- Visual Studio Code and Visual Studio
- Copilot CLI
- GitHub Copilot coding agent
- GitHub Copilot app
- JetBrains IDEs
- Xcode
- Eclipse
General availability does not mean every account receives access simultaneously. GitHub explicitly describes a gradual rollout, so an absent picker entry does not necessarily indicate a broken installation.
Business and Enterprise administrators can control Carbon through Copilot’s model policies. GitHub says new models are automatically enabled under default model enablement unless an administrator has disabled the global default or specifically disabled that model. For managed accounts, therefore, subscription eligibility is only one part of the access equation.
GitHub’s supported-model documentation also identifies Carbon as generally available, but its Carbon-specific minimum IDE and extension versions remain marked TBD. That is not confirmation that every older client supports it; GitHub recommends keeping the IDE and Copilot extension updated.
Usage-based billing deserves attention
GitHub says Carbon uses provider-list-price, usage-based billing. Its broader pricing documentation explains the mechanism: input, output, and cached tokens contribute to an interaction’s cost, which is converted into GitHub AI Credits at one credit per US cent. Included allowances vary by plan, and additional usage beyond those allowances is billable.
There is an important distinction between selecting a model for chat or agent work and receiving ordinary inline suggestions. GitHub lists Copilot Chat, CLI, and cloud agent among the features consuming AI credits. Code completions and next-edit suggestions are not billed in AI credits and remain unlimited on paid plans.
Agent work can also involve multiple model calls within one task. A request to investigate a cross-project failure may therefore consume substantially more usage than a short coding question—even when both begin with a single prompt. GitHub identifies task complexity, conversation length, and model choice as usage factors.
For individuals wanting to prevent additional charges, GitHub documents a specific control: set the Bundled AI credits budget to $0 USD and enable Stop usage when budget limit is reached. A paid subscription should not be mistaken for an unlimited model-usage tab.
A useful Windows development detail: billing display versions
GitHub’s billing guidance provides a practical reason to update Windows development tools beyond simply looking for Carbon. It warns that older clients can show incorrect model pricing, inaccurate usage information, outdated billing terminology, or missing usage alerts. Its recommended minimum versions for the usage-based billing experience include:
| Windows development tool | Billing-experience minimum |
|---|---|
| Visual Studio Code | 1.120 |
| Visual Studio 2022 | 17.14.33 |
| Visual Studio 2026 | 18.6.0 |
| Copilot CLI | 1.0.48 |
These are billing-experience recommendations, not confirmed Carbon compatibility minimums. Keeping those two categories separate avoids turning a documentation detail into an unsupported installation requirement.
Security-focused is not security-certified
Google positions Carbon for difficult, multistep coding and debugging, including extended reasoning and verification loops. Its launch explanation says Carbon uses more tokens than Gemini 4 Argon, while describing Argon as suitable for the majority of software-engineering and knowledge-work tasks. These are Google’s product claims, not independent measurements of performance inside Copilot.
The sensible evaluation is consequently task-specific: try Carbon on a bounded, difficult problem, inspect the proposed changes, run the relevant tests, and compare the result against its usage cost. That is an editorial recommendation—not a claim that Carbon has demonstrated a particular advantage.
Likewise, GitHub’s emphasis on secure code generation describes the model’s intended focus; it does not certify individual outputs as vulnerability-free. Treat Carbon as another engineering assistant to evaluate, not a reason to retire code review. The useful addition is choice—provided developers and administrators keep access, spending, and verification under deliberate control.