Call "Claude's flagship" right from the editor you already use — Opus 5 lands on 10 GitHub Copilot surfaces, and pricing shifts from multipliers to usage
Claude Opus 5, the flagship model Anthropic just released, is now available across 10 GitHub Copilot surfaces — VS Code, the CLI, cloud agents, and more. Pricing moves from multipliers to usage-based billing, organizations must explicitly enable it, and security-adjacent requests may be blocked.
Just days after Anthropic released its flagship model Claude Opus 5, it can now be used from the GitHub Copilot side as well. What stands out this time is that support isn't limited to a single IDE — it spans VS Code, the command line, and cloud agents all at once. Without switching your editor or development environment, you can now select the flagship Claude right from the model picker on your usual screen.
One model, available at every entry point to development
This time GitHub didn't confine Opus 5 to a specific IDE; it rolled the model out across all the major entry points developers touch. The supported surfaces are as follows.
- Visual Studio Code / Visual Studio
- Copilot CLI (command line)
- GitHub Copilot cloud agents
- GitHub Copilot app / github.com
- GitHub Mobile (iOS / Android)
- JetBrains / Xcode / Eclipse
It's available on Copilot's higher-tier plans — specifically Pro+, Max, Business, and Enterprise. The CLI added Opus 5 support in v1.0.75, so you can pick the same model from a terminal-resident agent too. Model mismatches like "the flagship in Claude Code, but a notch lower in Copilot" become far less likely, making it easier to keep the same brains across your tools.
Pricing is counted by "what you use," not by a "multiplier"
On the billing side, there's a clear difference from earlier Claude models. Opus 5 is billed on usage-based billing that tracks the provider's API list pricing, and the fixed multiplier of the old premium-request model no longer applies. Even when you're "using the flagship," how the cost appears changes.
| Method | How cost is determined |
|---|---|
| Old request multiplier (legacy) | Consumed at a fixed multiplier per request |
| Opus 5 usage-based billing | Billed by the amount of tokens actually consumed |
In other words, the longer an agent runs, the more directly consumption feeds back into the bill. On Business and Enterprise, members can't use it unless an administrator enables the Opus 5 policy in the Copilot settings. It's built so the organization can control who runs the flagship and when — one of the reins on "the token cost of keeping agents running" that we've touched on repeatedly on this blog. When rolling it out, the practical starting point is to enable it and draw the lines around who can use it.
A model tuned for long, autonomous work
GitHub positions Opus 5 for "complex, long-running coding work that demands careful reasoning, precise tool use, and reliable execution across multiple steps." In early evaluations, it showed its strengths in autonomous code changes, verifying regressions, and orchestrating multiple tools — and it was especially effective at targeted fixes, self-verification, and cutting wasted runs.
The "cuts wasted runs" assessment gains meaning when read alongside usage-based billing. Fewer moves tends to push token consumption down too, so it isn't just a matter of being smarter — it's a matter of operating cost. That said, this is based on GitHub's early testing, and the effect will vary with the codebase and task you're handling. The need to run it in your own environment and see for yourself hasn't changed.
The fine print behind the convenience
There are also caveats worth knowing before you adopt it. Opus 5 has hardened guardrails against high-risk cyber-related content built in, and these can block some security-adjacent requests. GitHub advises either rephrasing with legitimate context added, or choosing a different model.
In other words, even defense-oriented work such as vulnerability research, penetration testing, or analyzing attack code may be rejected depending on how it's phrased. Erring on the side of safety is reasonable design in itself, but for developers doing security work it can be a source of friction. It's safer for the team to decide in advance on a procedure for switching to an alternative model, on the assumption that these misses will happen. It's best to view convenience and constraint as arriving together, two sides of the same coin.
What to check right now
To sum up: while you can now call the flagship Claude straight from your everyday development environment, the practical points to keep in mind are three — cost accrues by the amount you use, organizations need explicit enablement, and security-adjacent requests can be rejected. If you're on a higher-tier plan, first check the policy's state in the admin settings and make sure usage-based consumption is visible. From there, the shortcut to measuring the effect while avoiding runaway spending is to run a small test of just how much "wasted execution" drops on long-running tasks.
References: Claude Opus 5 is now available in GitHub Copilot (GitHub Changelog) / Models and pricing for GitHub Copilot (GitHub Docs) / Claude Opus 5 (Anthropic)
