GitHub announced on September 22, 2026, that Claude Opus 5.5 is available in GitHub Copilot. The model is aimed at long-running agentic coding tasks and knowledge work, giving eligible Copilot users another option for repository-scale changes, multistep work, and tasks that require recovery from errors.
Access is not necessarily immediate for every eligible account. GitHub says the model is rolling out gradually, and availability can depend on the user’s plan, Copilot client, and organization settings.
What changed in GitHub Copilot
Claude Opus 5.5, described by GitHub as Anthropic’s newest Opus model, has been added to the Copilot model catalog with general availability status. GitHub’s model-comparison guidance recommends it for long-running agentic coding and knowledge work, particularly where efficient multistep execution, error recovery, and collaboration are important.
In its announcement, GitHub said early testing found that Opus 5.5 resolved tasks comparably to Claude Opus 5 while using significantly fewer steps and tokens. GitHub also reported that the model recovered quickly from errors during multistep tasks. The announcement does not provide benchmark methodology or quantified savings, so those statements should be treated as GitHub-reported testing results rather than an independently verified performance comparison.
Claude Opus 5.5 can be selected through the model picker where it is available. The announcement lists Visual Studio Code, Visual Studio, Copilot CLI, GitHub Copilot coding agent, the GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode, and Eclipse as supported selection surfaces.
Who can use Claude Opus 5.5
According to GitHub’s announcement, the eligible plans are Copilot Pro+, Max, Business, and Enterprise. Model availability can still vary by plan and by where Copilot is being used, and GitHub’s supported-model documentation notes that availability may change over time.
Business and Enterprise administrators can manage access through Copilot model policies. Under GitHub’s default model-enablement behavior, new models are enabled unless an administrator disables the global default or disables the specific model. Organizations that do not want the model available should therefore review their model-policy settings during the rollout.
The supplied documentation does not establish a complete client-by-client entitlement matrix, a geographic rollout scope, or a final date by which every eligible account will receive access. If the model is missing from the picker, that does not necessarily indicate a configuration problem.
Pricing and Copilot credit considerations
GitHub’s supplied pricing documentation lists Claude Opus 5.5 at $4.00 per 1 million input tokens, $0.20 per 1 million cached input tokens, $5.00 per 1 million cache-write tokens, and $20.00 per 1 million output tokens. The documentation says Copilot usage consumes input, output, and cached tokens, and that the interaction cost depends on the selected model and the tokens consumed.
One GitHub AI credit equals $0.01 USD. Usage beyond included plan allowances is billed in GitHub AI credits at the listed per-token rates. The plans documentation lists monthly allowances of 7,000 AI credits for Copilot Pro+, 20,000 for Copilot Max, 1,900 per user for Copilot Business, and 3,900 per user for Copilot Enterprise. Actual usage can therefore depend on both the model’s token consumption and the applicable plan allowance or organizational policy.
Long generated responses may have a greater effect on usage than input alone because the listed output-token rate is higher than the input and cached-input rates. Developers evaluating Opus 5.5 for extended agent tasks should monitor usage rather than assume that included credits will cover every workflow.
GitHub’s plan documentation separately says that code completions and next-edit suggestions are not billed in AI credits. That treatment should not be generalized to all Copilot interactions, which remain subject to the token and plan rules described in the pricing documentation.
Where the model may fit
Claude Opus 5.5 is positioned for work that continues across multiple steps rather than only short, isolated prompts. Examples supported by GitHub’s published guidance include long-running agentic coding, knowledge work, multistep tasks, error recovery, and collaboration.
That positioning may make the model worth considering for changes that require an agent to inspect context, make several related edits, respond to failures, and continue toward a result. It does not establish that Opus 5.5 is universally better than Claude Opus 5 or other Copilot models. Workload characteristics, token consumption, plan limits, and the surrounding Copilot surface all remain relevant.
What developers and administrators should do
- Check the model picker on the Copilot surface you use, including your IDE, Copilot CLI, coding agent, or GitHub web and mobile experience.
- If Claude Opus 5.5 is not listed, check again later because GitHub is using a gradual rollout.
- If you use Copilot Business or Enterprise, review the organization’s model policies and default-model settings.
- Before assigning long-running or token-intensive work, review the applicable AI-credit allowance and the model’s listed token rates.
- Evaluate GitHub’s efficiency claims as vendor-reported results, and measure the model against your own tasks before changing standard workflows.
The supplied sources do not document a Claude Opus 5.5-specific API identifier, CLI command, or configuration-file setting. Selection through the model picker is the documented configuration path.
Watermarking and its limitations
Claude Opus 5.5 outputs are watermarked. GitHub says the watermark does not change meaning, quality, or readability and does not add tokens or cost.
Anthropic’s explanation of the text watermark says it is not visible or distinguishable to readers, adds no hidden characters, and carries no identifying information. It is intended to indicate the likelihood that Claude was involved in producing or processing text.
The watermark should not be treated as proof of authorship, ownership, or legal responsibility. Anthropic also says detection is less effective on small samples and can be weakened by substantial rewriting. It is therefore a probabilistic signal, not an authorship guarantee or a security control.
Availability caveats
Claude Opus 5.5 is listed as a generally available supported model, but the rollout itself is gradual. Account-level, geographic, and exact client-by-client availability details are not fully documented in the supplied sources. GitHub also notes that supported-model availability can change over time.
For now, the practical test is whether the model appears in the Copilot model picker for an eligible account. Administrators should also account for organization policy, while individual developers should confirm pricing and credit implications before using Opus 5.5 for extended agent sessions.
