GitHub has introduced team-level customisation for Copilot, allowing organisation owners and team maintainers to set default Agent Plugins across their groups. The feature, now generally available for Copilot Business and Enterprise customers, streamlines workflow setup by eliminating repetitive individual configurations. Alongside this, GitHub expanded Copilot Fusion’s capabilities, adding new models and reporting strong benchmark results on Terminal-Bench 4.1.
Team plugins simplify onboarding
Team maintainers can now define approved Agent Plugins as defaults for their members. These settings apply automatically in VS Code, the GitHub Copilot app, and Copilot CLI. Members retain control and can disable team plugins where policy allows. The feature operates within existing enterprise and organisational policies—required or blocked plugins still take precedence, and only direct team members inherit the defaults, not child teams.
To set up, maintainers navigate to the team page on GitHub, select the ‘Customize Copilot’ tab, and browse plugins from integrated marketplaces or add custom ones from a repository. This reduces friction in onboarding and ensures consistency in tooling without enforcing rigid central mandates.
Copilot Fusion advances in performance and access
Copilot Fusion has been added to the public Terminal-Bench 4.1 leaderboard, an independent benchmark for AI agents performing complex terminal tasks. Copilot Fusion Max achieved a 71.2% resolution rate—5.4 percentage points ahead of Opus 5.5 and Fable 5.2—while costing 88% less than Fable 5.2 and 57% less than Opus 5.5. These figures position it as a cost-efficient option for high-complexity automation.
Fusion is now available on github.com, Visual Studio, and the GitHub Mobile beta (iOS and Android), with production rollout imminent. Users select Fusion in the model picker, set an effort level, and let the system coordinate multiple models per task.
New models enhance tiered performance
Three new models have been integrated into the Fusion tiers:
- Claude Sonnet 5.5 joins Copilot Fusion Max
- GPT-6.1 Terra is now in Copilot Fusion High
- MAI-Code-2 powers Copilot Fusion Standard
This expansion allows finer alignment between task complexity and model capability. The tiered structure ensures teams can balance cost, latency, and accuracy across different workflows.
What it means
For UK and UAE tech teams, this update reduces setup overhead and strengthens governance. Organisations in regulated sectors—such as finance in London or government tech in Abu Dhabi—can now enforce baseline tooling while preserving team autonomy. In Kazakhstan, where software teams often rely on lean infrastructure, the cost-performance edge of Copilot Fusion Max could be decisive.
The Terminal-Bench results offer an objective benchmark, rare in AI tooling—this helps IT leads compare Copilot against alternatives without relying solely on vendor claims. The local execution model of Fusion also reduces data exposure compared to fully cloud-dependent agents.
Still, dependency on GitHub’s ecosystem remains high. While team plugins improve control, they deepen integration with GitHub’s proprietary services. Organisations seeking vendor neutrality may prefer open alternatives like the Blender AI Copilot extension, which supports local models via Ollama and OpenRouter and exposes its MCP server for cross-tool use.
GitHub’s move signals a shift toward structured AI governance at the team level. For founders and IT leads, the priority should be assessing whether Copilot’s closed ecosystem aligns with long-term data strategy—or if modular, open alternatives offer better control.