Skip to content
NewsProduct launch

GitHub Copilot adds Claude Opus 5.5 for agentic tasks and ships JetBrains 1.18.0 with org-wide shared skills

· by Pondero Newsdesk

The short version

GitHub's September 22 changelog delivered two updates: Claude Opus 5.5 is now selectable across every Copilot surface for agentic sessions, and JetBrains 1.18.0 introduces org-wide shared skills and instructions alongside AI-assisted tool approvals.

GitHub Copilot adds Claude Opus 5.5 for agentic tasks and ships JetBrains 1.18.0 with org-wide shared skills

GitHub published two Copilot updates on September 22, 2026: Claude Opus 5.5 is now available across every Copilot surface for agentic coding and long-running tasks, and version 1.18.0 of the JetBrains plugin introduces org-wide shared skills and AI-assisted tool approvals.

Claude Opus 5.5 lands in Copilot

Anthropic's newest Opus model is now selectable in the model picker across VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, github.com, GitHub Mobile, and the Copilot coding agent, per the GitHub Changelog. Access is open to Pro+, Max, Business, and Enterprise subscribers.

In early testing, Opus 5.5 resolved coding tasks at a comparable rate to Opus 5 while using significantly fewer steps and tokens, and recovered from errors more quickly in multistep flows. GitHub positioned it as the recommended Opus-class model for Copilot agent sessions. Billing runs at provider list pricing under usage-based billing. Rollout is gradual.

One operational detail worth flagging: Opus 5.5 applies Anthropic's text watermarking to outputs. Per GitHub's announcement, the watermark does not change meaning, quality, or readability, and adds no tokens or cost.

What JetBrains 1.18.0 adds for enterprise teams

The change that matters most for engineering org administrators is shared skills and instructions, per the JetBrains changelog. Local and agent sessions now support organization and enterprise skills alongside org-managed custom instructions. A Copilot admin can publish coding conventions, tool permissions, or domain context once and have every developer in the org inherit them automatically, without per-user setup.

AI-assisted tool approvals in public preview are another addition: low-risk tool calls in agent sessions receive automatic approval, while higher-risk actions still prompt the developer for a decision. The practical effect is fewer approval interruptions during long agent runs without surrendering control over consequential steps.

Two further additions round out the release. Re-editing an earlier message now rewinds both the conversation and file changes from that point, which cuts the cost of correcting early misdirection without stacking another message on top. Codex agent gains plan mode, letting developers review and approve a proposed approach before implementation begins.

Admins can now publish one skill set org-wide

For a team already on Copilot Business or Enterprise, the org-wide skills feature changes the overhead equation. Before 1.18.0, consistent agent behavior across an engineering team required each developer to maintain their own instructions and skill configuration. One admin-published set now propagates automatically. That is a direct argument in procurement conversations comparing Copilot against per-developer tool configurations.

The Opus 5.5 model addition matters on a separate axis. Enterprise teams running long-horizon coding agents on Copilot were previously ceiling-constrained by Opus 5, which is capable but token-heavy across multistep flows. Fewer steps and faster error recovery mean shorter agent sessions at comparable output quality. Whether that translates into lower per-task cost depends on Microsoft's billing model for Copilot Business and Enterprise, which has not been updated to reflect Opus 5.5 compute costs as of September 24, 2026.

Cursor's model picker availability for Opus 5.5 is a secondary signal to track, given the overlap in enterprise agentic coding workflows between the two products.

Sources