Updated on
GitHub Copilot: reviews and analysis
The coding assistant built into GitHub and your IDE, from $10.
Affiliate link · no extra cost to you
Our verdict
Copilot is the lowest-friction path to putting AI in the workflow: it lives in the IDE you already use and in GitHub, where it reviews pull requests and takes tasks as an agent. Its free plan and $10 entry tier make it the most accessible in the category. On agent autonomy it trails the specialists, and plan credits drain fast on premium models.
Best for: Teams living in GitHub who want AI without changing tools.
What changed this week
Sweep of
Up a tenth: sign-ups for the individual plans, paused since 1 June, were confirmed reopened since 17 June (realPrice up from 3.0 to 3.2), though the 31 July removal of Gemini 2.5 Pro and Gemini 3 Flash weighs a little on support.
- down
- up
Up on the latest sweep after moving the other way.
We have tracked it since July 25, 2026: 3 sweeps on record, with the score moving between 3.8 and 3.9. We publish the latest change here; the full series is not published.
What the internet says
Base prices have not moved: Free at $0, Pro at $10, Pro+ at $39 and Max at $100. What this sweep corrects from the previous one matters: sign-ups for the individual plans, paused since 1 June, reopened gradually from 17 June, and today the pricing page shows no waitlist at all. What has not improved is the pace of model cuts, on 31 July GitHub retired Gemini 2.5 Pro and Gemini 3 Flash effective immediately and announced further removals, including both Opus models, for 1 September. And the underlying complaint is the same one from June, AI credits run out before month end.
What the web repeats in favour
- It lives where you already work: your usual IDEs, pull request review and agent tasks inside GitHub
- A cheap front door that is open again: there is a real free plan, Pro is still $10 a month, and the sign-ups that had been paused since June no longer are
- The model catalogue keeps growing, with Kimi K2.7 Code generally available since July 2026
What the web repeats against
- AI credits run out well before month end under heavy use, and it is the most repeated complaint since June 2026, with new August threads from users burning through half their quota in two days
- There is no way to estimate a task's cost before running it, so the bill shows up after the fact
- The model catalogue changes without much notice: on 31 July, Gemini 2.5 Pro and Gemini 3 Flash were retired effective immediately, with more removals announced for 1 September, including both Opus models
- Anyone still on a legacy annual plan does not get new models or features at the same pace as current plans
Sweep sources: Official pricing · Press · Docs · Press · Press · Communities · Communities
Latest GitHub Copilot news
Copilot's impact dashboard adds a return-on-investment section
The new section compares cost per developer and percentage of payroll against pull requests produced, but GitHub warns the cost figures are estimates based on AI credit consumption and the salary selector is a modeling input, not actual payroll data. Source
Copilot adds the Kimi K3 model with its own per-token rate
Kimi K3, an open-weight model GitHub hosts on Fireworks AI, lands on Copilot's Pro, Pro+, Max, Business and Enterprise plans at $3 per million input tokens and $15 per million output tokens. On Business and Enterprise it ships off by default, and an admin must enable the policy before anyone can use it. Source
Copilot lets Enterprise accounts allowlist which MCP servers can run
The new allowedMcpServers and deniedMcpServers keys in managed settings let an Enterprise admin approve or block specific MCP servers across the Copilot app, CLI and VS Code, with a policy that blocks any malformed configuration by default. Source
Copilot cloud agent adds a reasoning-level selector per task
When delegating a task to the cloud agent, users can now set how much the model reasons before responding, alongside the model choice. Raising the level improves results on complex problems but burns more tokens and credits, and lands on Pro, Pro+, Business, Enterprise and Max plans. Source
Copilot cloud agent can now be triggered by issue and PR comments
Repository owners can configure automations from the Agents tab that fire when a specific comment text is posted, for example to update docs or open tech-debt issues. Available on Pro, Pro+, Max, Business and Enterprise, and on the latter two an admin must first enable the cloud agent policy. Source
Pros / Cons
Pros
- Built into GitHub and the usual IDEs
- Free plan and a $10 entry tier
- PR review and an agent on the platform
Cons
- Less agent autonomy than the specialists
- Premium credits run out fast
TLDR: Copilot is the lowest-friction path to putting AI in your flow: it lives in the IDE you already use and in GitHub, where it reviews pull requests and takes tasks as an agent. Free plan with limited completions and a $10-a-month paid entry, the category’s cheapest. On agent autonomy it trails the specialists, and premium credits drain faster than they look.
What GitHub Copilot is and how it works
Copilot invented the AI autocomplete category and has grown with it: today it is a complete assistant with editor chat, an agent mode for tasks, pull request review and the ability to assign it work directly from GitHub, like one more collaborator on the repository.
Its structural advantage is not technical but positional: it is where you already work. In the IDE (VS Code and the usual ones) as an extension, and in GitHub as part of the platform, so the assistant sees issues, pull requests and project context with zero configuration. For a team living in GitHub, enabling it is a one-click decision.
The pricing system combines subscription and credits: autocomplete is unlimited on paid plans, while chat, agent and premium model features consume monthly credits by usage.
What it is like day to day
Autocomplete remains the product’s heart and what it does best: line and block suggestions that hit often enough for your fingers to learn to tab. For anyone writing code daily, that zero friction is the argument.
Chat and the agent cover the middle ground: explaining code, generating tests, bounded tasks. The platform agent, which takes an issue and returns a pull request, works well on small, well-defined tasks and falls short on the long, ambiguous ones where Claude Code or Codex show more judgment. Automatic pull request review, meanwhile, is among the package’s most useful pieces: it catches the mechanical before a human spends time.
The real friction is credits: premium models drain them fast, and the jump from Pro to Pro+ exists precisely because the entry plan runs short if you use the agent seriously.
Pricing and plans
The free plan gives a limited number of completions and basic access, enough to taste the experience. Pro costs $10 a month with unlimited autocomplete and monthly credits for chat and agent. Pro+ climbs to $39 with premium models and more credits, higher tiers exist for intensive use, plus the enterprise plans.
It is the category’s cheapest entry, and that is its card: for $10 you get unlimited autocomplete and a decent agent. The comparison with Cursor, which costs double and delivers more agent, is developed in the Cursor vs Copilot comparison.
Who it is for (and who it is not for)
Copilot is for teams and developers living in GitHub who want AI without changing tools: autocomplete in the usual IDE, PR review on the usual platform. It is also the sensible entry option for anyone unsure how much AI they will use: starting at $10 and measuring beats starting at $40.
It is not for anyone after maximum agent autonomy, which is terminal specialist territory. Nor for anyone wanting the editor rebuilt around AI, which is Cursor’s and Devin Desktop’s (the former Windsurf) offer.
GitHub Copilot alternatives
Cursor is the classic comparison: double the price for more agent and more integration. Claude Code and Codex play another autonomy league from their assistant subscriptions. And for pull request review as a specialty, CodeRabbit goes deep where Copilot generalizes. The full list is at GitHub Copilot alternatives and the ranking in the best AI coding tools.
Frequently Asked Questions
Is the free plan enough to work with?
It is enough to try: the monthly completion cap drains fast under daily use. Real work starts at Pro, which at $10 is the category’s cheapest entry.
What are premium credits?
The currency paying for chat, the agent and frontier models: each plan includes a monthly pool and intensive use drains it. Autocomplete consumes no credits on paid plans.
Does Copilot’s agent compete with Claude Code?
On small, well-defined tasks, it delivers. On long, ambiguous tasks or complex codebases, the specialists hold a clear lead in autonomy and judgment.
Is it worth it if I do not use GitHub?
It loses its positional advantage: without the platform integration what remains is good autocomplete with a decent agent. In that case, compare directly against Cursor and the terminal agents before deciding.