GitHub Copilot CLI
Models from Anthropic, OpenAI, Google and Microsoft in one subscription; Copilot Fusion picks the model, and a task moves between CLI, web and phone.
Price and access
- Free?
- A free plan with limits
- Price
- Free / Pro $10 / Pro+ $39 / Max $100, usage in AI credits
- From Russia
- GitHub's trade-control rules bar selling Copilot to Russia and Belarus.
- Models
- Anthropic · OpenAI · Google · Microsoft MAI, автовыбор Copilot Fusion
- Licence
- not disclosed
- Our benchmark
- Claude Sonnet 4.6: 8.18 of 10, rank 15 →
Unavailable from Russia. Similar tools that work: Kimi Code CLI, ZCode, DeepSeek Harness.
What changed in price and access
- GitHub Enterprise Cloud now includes Copilot access with no bundled credits: usage is metered from the first request; self-serve comes later ↗
- GitHub Copilot Automations can now be capped in AI credits per run, day, week or month; shared automations are billed to the repository owner ↗
- Copilot Fusion is generally available on all plans, including Copilot CLI: it routes between models, with a temporary 30% discount on model costs ↗
- Copilot cloud sessions run in GitHub sandboxes; from November 2026 their compute is billed against the GitHub Actions entitlement ↗
- GPT-6.1 Terra arrived in GitHub Copilot, a cheaper model for agentic coding and terminal work ↗
Changes for this tool in total: 14. The full feed for all agents →
What it gives a manager
- Automations: on a schedule or event (new PR, security alert) the agent writes a digest or fixes a bug; credit caps per run, day or month
- Shared automations belong to the repository: they keep running after the author leaves the team, and the organisation pays
- Budgets per person and cost centre, e.g. $20 per engineer and $5 for marketing, with a hard stop when they run out
- Start a task from your phone, Slack or Teams and pick up the result in the CLI or on the web: it is one and the same session
Weak points
- In Microsoft's own demo the ToDo Sweeper app built in 20 minutes did not work, and the speaker ended the talk without a fix.
- Since 2026-08-20 billing is per token in AI credits and annual plans are gone; from November cloud sandboxes also draw on Actions quota.
- Autopilot grants every permission and loops until it decides it is done; the speaker recommends it only in a sandbox.
- The instructions file and MCP servers fill the context window on every launch; unused MCPs have to be disabled by hand via /mcp.
How to check this agent’s work
- Copilot review resolves its own comments: check that the fix was actually made.
- Cap the AI credits per automation: an extra run shows up at once.
- Give a disputed task to two models: Copilot switches between them in one window.
Picking an agent takes an evening. After that the question is not the price but whether you can trust what it did.
You’re on the list
We will write when enrolment opens. Meanwhile, here is what the course covers.
The short course programmeWhat the course covers: programme and price
Strengths
- Models from five vendors in one window; Copilot Fusion picks and combines them, for now with a 30% discount on model costs.
- Pro at $10 a month undercuts Codex, Claude Code and Cursor, and includes access to the Claude Code and Codex agents (Byte & Beyond).
- Rubber Duck: code written by Claude Haiku 4.5 was reviewed by a different vendor's model, GPT-5.6 Terra, in Microsoft's live demo.
- Plan mode (Shift+Tab) asks clarifying questions and writes plan.md without touching code; /fleet hands plan items to subagents in parallel.
Alternatives
Kilo Code
ZCode
Kimi Code CLIWho it is for
- Team already living in GitHub and a Copilot subscription
- [object Object]
Example from a review: A TODO scanner in 20 minutes on the VS Live stage
Kayla, a Senior Dev Advocate at Microsoft, spent 20 minutes building ToDo Sweeper in Copilot CLI, a scanner for TODO and FIXME comments in a repository. She switched to Claude Haiku 4.5 for speed, enabled YOLO, ran /init, answered Plan mode's clarifying questions, then launched Autopilot and /fleet so the plan's eight items ran in parallel. The build finished, but the scanner hung on the Windows Terminal repo, a rebuild did not help, and it was still broken when the talk ended. /diff, /review and Rubber Duck were shown on the broken code.
Source: Microsoft Developer, "Fast Focus: Dream a Little Dream: Dev with the CLI"
Reviews (2)
The video authors’ opinions, not our conclusions.
- Fast Focus: Dream a Little Dream: Dev with the CLI · Microsoft Developer, 18 August 2026A Microsoft speaker presents Plan, Autopilot, /fleet, /context and Rubber Duck as a complete terminal toolkit; the demo app itself did not work.
- Codex vs Claude Code vs Cursor vs Copilot – Best AI Coding Agent in 2026? · Byte & Beyond, 31 August 2026Without running any code, the author sees Copilot as the path of least resistance for GitHub-centric teams: $10 for Pro and several models and agents in one subscription.
Facts with sources
- Opened to all Copilot subscribers on February 25. Source (in Russian)
- Purchased by 56% of teams, used daily by 14% (LeadDev, 2026). Source (in Russian)
- A budget approved in Q1 at the rates of the time was outdated by June because the pricing model changed. Source (in Russian)
- GitHub's trade controls explicitly bar selling Copilot to Russia and Belarus; location is judged by IP and payment history. Source
- Plans: Free, Pro $10, Pro+ $39 and Max $100 a month; 1 AI credit = $0.01, and Pro includes $15 of credits. Source
- Since 28 October automations can be capped in AI credits per run, day, week or month. Source
