Microsoft / GitHub's multi-model agentic terminal. The only major coding CLI that switches between Claude (Sonnet/Opus 4.6) and GPT-5.2–5.5 from one tool, via /model. Plan and Autopilot modes, worktree isolation, self-iterating code review, specialised Explore and Task sub-agents. Bundled with Copilot Pro / Pro+ / Business / Enterprise; moves to usage-based AI Credits June 1 2026.
Copilot CLI is GitHub's response to "agents should live in the terminal." Tightly integrated with GitHub repos, PRs, and Actions; multi-model by default; isolation-first; designed for teams already on the Microsoft / GitHub side of the ecosystem.
Where Claude Code commits to Anthropic and Codex CLI commits to OpenAI, Copilot CLI deliberately doesn't pick. As of May 2026 it supports Claude Sonnet 4.5 (default), Sonnet 4.6, Opus 4.6, GPT-5.2-Codex, GPT-5.3-Codex, GPT-5.4, and GPT-5.5. Switch models mid-task with /model. For teams hedging on model choice — or wanting to route deep reasoning to Claude and fast iteration to GPT-5-mini-class — this design is structurally different.
The CLI offers two execution modes: Plan mode (approval-by-default, the agent proposes before doing) and Autopilot mode (hands-off autonomy, the agent iterates without stopping). Plus two isolation strategies: worktree isolation creates a separate Git worktree so the agent's changes don't touch your branch until you review; workspace isolation applies changes directly for faster iteration. Both are configurable per-task.
Specialised sub-agents come built in: Explore for fast codebase analysis, Task for running tests and builds. You can define custom agents in ~/.copilot/agents/.agent.md. Background delegation via /delegate lets the agent work while you keep typing.
Six load-bearing capabilities. Verified against current docs.
Claude 4.5/4.6, Opus 4.6, GPT-5.2–5.5 from one CLI. /model mid-task. The only major coding CLI that does this structurally.
Plan: approval-by-default, agent proposes, you accept. Autopilot: hands-off iteration through the task. Switch per task.
Worktree creates separate Git worktree so changes don't touch your branch until reviewed. Workspace applies directly for fast iteration. Configurable.
Built-in: Explore (codebase analysis), Task (running tests/builds). Custom agents in ~/.copilot/agents/.agent.md. /delegate for background work.
Agent reviews its own changes via Copilot code review before opening a PR, gets feedback, iterates. You don't see broken PRs.
Diagnose issues from bug-report screenshots. JetBrains IDEs can delegate tasks to a locally running Copilot CLI agent (May 2026 preview).
Tighter coupling to GitHub than the other two CLIs. Sits naturally next to your repo, PRs, Actions, and Copilot Workspace.
copilot command launches the agent. Linux, macOS, Windows (PowerShell v6+).
Native integration. Same agent, different surface. Agent mode in VS Code is the gateway most teams encounter first.
Public preview May 2026. JetBrains IDEs delegate work to a locally running Copilot CLI agent. Brings the CLI's depth into IntelliJ / GoLand / PyCharm without re-implementing.
Separate browser-based collaborative environment. Different product surface but same Copilot ecosystem.
Copilot historically was a flat-rate per-seat subscription. From June 1 2026, the model shifts to usage-based AI Credits — tokens converted to credits at per-model rates, deducted from a monthly allotment.
Copilot Pro — $10/month + $10 in monthly AI Credits. Entry tier; cheapest of the three major CLIs.
Copilot Pro+ — $39/month + $39 in monthly AI Credits.
Copilot Business — $19/user/month + $19 in monthly AI Credits.
Enterprise — custom pricing.
Usage-based billing: 1 AI Credit = $0.01 USD. Tokens consumed (input, output, cached) convert to credits at per-model rates. GitHub Copilot CLI is included as part of the subscription and consumes the monthly Credit allotment.
The June 2026 move makes Copilot the only major coding CLI on true usage-based billing. Claude and OpenAI gate by tier; GitHub gates by tier plus a per-credit meter that converts tokens to dollars.
Implications:
— Low-usage teams may pay less under credits than under flat-rate.
— Heavy-usage teams may pay more — especially when calling Opus 4.6 or GPT-5.5 (higher per-token cost → faster credit burn).
— Cost-modelling matters more than under flat-rate tiers. Budget per developer per month based on expected token volume.
Honest tradeoffs. Multi-model and GitHub-native are its strengths; usage-based pricing is the surface to model carefully.
Pricing. Copilot Pro at $10/month + $10 credits is the cheapest entry of the three major CLIs — roughly R190 + R190 credits per developer per month at mid-2026 rates. For a 10-person team that's around R45k/year, similar to Claude Pro but with the multi-model flexibility built in.
GitHub gravity. GitHub is the dominant SCM in SA dev shops — banks, telcos, SMEs, the public-cloud-native crowd. That alone gives Copilot CLI a default-pick gravity for many teams. The friction of "set up GitHub auth" is zero (already done); the friction of "set up a third subscription" is real.
Multi-model as a hedge. Model preferences shift — the model that's best for your work in May 2026 may not be best in November 2026. Copilot CLI's multi-model design hedges that bet for free. Run Claude for deep reasoning tasks, GPT-5-mini for fast iteration, switch as benchmarks shift.
Usage-based credits and budgeting. The June 2026 move to usage-based credits requires more thoughtful budgeting than flat tiers. Track per-developer credit burn for the first month and adjust the tier choice. Model carefully before committing a large team.
POPIA. Microsoft Azure has SA regions and offers data-residency commitments at the Enterprise tier. For regulated industries this is often the deciding factor.