GitHub made Claude Opus 5 available in GitHub Copilot on 24 July 2026. The announcement is short, and three details in it matter more than the headline: which plans actually get it, how it is billed, and a safeguard that can block security work. If you work in security, the third one will affect you before the first two do.
Who actually gets it
GitHub lists Claude Opus 5 for Copilot Pro+, Max, Business and Enterprise users.
That is worth comparing against the Sonnet-class release. When Claude Sonnet 5 became generally available in Copilot, GitHub listed it for Copilot Pro as well. Opus 5 starts one tier higher. If you are on a plain Pro subscription and expecting to find it in your model picker, you will not, and it is not a rollout delay.
On Business and Enterprise, availability is not automatic either: an administrator must enable the Claude Opus 5 policy in Copilot settings first. If your colleagues can see it and you cannot, that is the setting to ask about.
The rollout is described as gradual, so a qualifying plan still does not guarantee the model appears immediately.
The billing detail people will miss
GitHub states that Claude Opus 5 is billed at provider API list price under usage-based billing.
Read that carefully, because it is not the model most Copilot users have internalised. Much of Copilot's premium model access works through a premium request multiplier: each request counts as a fixed number of premium requests against your allowance. Billing at API list price under usage-based billing instead means your cost tracks actual token consumption.
The practical consequence shows up in agentic work, which is precisely what this model is positioned for. A long autonomous session that reads many files, runs tools and iterates will consume tokens in a way that a per-request count would have flattened. That is not a criticism, it is a different cost shape, and it is worth knowing before you point it at a large refactor.
The safeguard that matters if you do security work
This is the part of the announcement most coverage will skip, and it is the most consequential for this site's readers.
GitHub states that the model includes enhanced safeguards for high-harm cyber content, and that these safeguards may block some cyber-related or security-adjacent requests.
"Security-adjacent" is doing a lot of work in that sentence. If your job involves malware analysis, exploit development, offensive testing, or simply reasoning about how an attack works in order to defend against it, you should expect legitimate professional requests to be refused some of the time. That is a documented property of the model as shipped, not a misconfiguration to file a bug about.
The honest practical advice: if security work is your daily use case, do not switch your default model to Opus 5 without testing it against your actual prompts first. Keep another model available for the requests it declines, and find out where the boundary sits for your work rather than discovering it mid-task.
What it is positioned for
GitHub describes strong performance on agentic coding workflows, specifically naming autonomous code changes, regression verification, and tasks that require coordinating multiple tools.
That is a coherent picture: this is aimed at the multi-step work where the model plans, edits, checks its own output and uses tooling, rather than at single-shot completions. Whether it is worth the billing difference for your workload is an empirical question about your own repositories, and we have not benchmarked it.
Available surfaces are broad: Visual Studio Code, Visual Studio, Copilot CLI, the GitHub Copilot cloud agent, the GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains, Xcode and Eclipse.
The bottom line
Claude Opus 5 landed in GitHub Copilot on 24 July 2026 for Pro+, Max, Business and Enterprise, one tier above where Sonnet 5 started, with administrator enablement required on the organisation plans.
Two things to internalise before you use it: it is billed at provider API list price under usage-based billing, so agentic sessions cost what they consume, and it carries safeguards that may block security-adjacent requests, which is a real constraint if that is your field rather than an edge case.
Based on GitHub's published changelog announcement of 24 July 2026. Plan names, the billing statement, the supported surfaces and the safeguards wording are GitHub's, not our characterisation. We have not benchmarked this model against others and quote no performance figures of our own. Model availability and pricing change frequently, so verify the current state for your plan.
