GitHub Copilot Pricing: Which Tier Do You Actually Need?

GitHub Copilot's four tiers, as of August 2026

GitHub Copilot's individual pricing is Free, Pro, Pro+ and Max, verified directly against github.com/features/copilot/plans in August 2026. Two organisation-level plans, Business and Enterprise, sit above those four but aren't priced publicly the same way.

The rest of this article goes through what each tier's own wording actually promises, since "unlimited" and "premium models" hide real differences in what you can do day to day, and closes with which tier you need if what you're actually trying to do is run agent skills or hand work off to Claude Code or Codex from inside Copilot.

This is worth working through carefully rather than defaulting to whichever tier looks like the obvious upgrade, because the tiers don't scale the way a lot of SaaS pricing does. Pro to Pro+ isn't "the same features, more of them", it gates a genuinely different workflow, delegating a task to a background agent instead of chatting with one. Pro+ to Max, by contrast, gates nothing new at all. Picking based on price alone in either direction risks either paying $39 for a feature you'll never use, or staying on $10 and wondering why delegation never appears in your Copilot chat window.

Free: what's actually included

Free costs nothing and gives you 2,000 completions a month, access to a stated set of models including Haiku 4.5 and GPT-5 mini, Copilot CLI, and community support. What's easy to miss is that Free also lists agent mode support, custom instructions and agents, and MCP server integration as included, not gated behind a paid tier.

That matters specifically for agent skills. A SKILL.md folder committed to .github/skills/ in your repository:

.github/
  skills/
    commit-message-style/
      SKILL.md

works with Copilot's own default assistant in agent mode on the Free tier exactly as it does on Pro, Pro+ or Max. Nothing about reading and activating a skill is paywalled. What Free actually limits is volume (2,000 completions) and which models you can pick, not whether agent mode or Agent Skills work at all. See how to install agent skills in GitHub Copilot for the full setup.

Pro ($10/month): the tier most individual developers land on

Pro adds unlimited code completion and next-edit suggestions, model selection, cloud agent and code review access, the ability to assign work to Copilot for pull requests, app modernisation tooling for Java and .NET, and $15 of monthly AI credits on top of everything Free includes.

The detail worth calling out specifically: Pro's own bullet list includes "access to 3rd party agents (Claude Code and Codex)". That means you can pick Claude or Codex as the model actually answering you inside Copilot's chat and agent mode on the Pro tier, you aren't limited to GitHub's own assistant once you're a paying customer. What Pro does not include is delegation, more on that distinction below.

For most individual developers who've outgrown Free's completion cap but don't need premium models specifically, Pro is the tier that matches actual daily use: writing code, running agent mode against a codebase, having Copilot open pull requests, all without hitting a ceiling that forces an upgrade.

Pro+ ($39/month): where delegation to Claude and Codex unlocks

Pro+ adds premium model access including Opus, audit logs, 4x or more included usage compared to Pro, and $70 of monthly credits. The feature that actually separates Pro+ from Pro, rather than just being "more of the same", is delegation: Pro+'s bullet reads "delegate tasks to third-party coding agents like Claude by Anthropic and OpenAI Codex."

That's a different capability from Pro's "access to" Claude and Codex. On Pro you choose Claude or Codex as the model driving an interactive session you're actively steering. Delegation is the assign-and-walk-away workflow: hand a task to Claude or Codex as a background coding agent, the same shape as GitHub's own Copilot coding agent, and it works independently and opens a pull request for you to review. GitHub's rollout of third-party agents on Copilot describes that delegation flow as available in preview specifically for Pro+ and Max, not for Pro, which lines up with how the two tiers' own bullet points are worded on the plans page.

If your actual use case is "let Claude Code work through a ticket while I do something else," that's a Pro+ feature, not a Pro one, regardless of whether you can already chat with Claude interactively on Pro.

Worth knowing if you're on a Business seat rather than an individual one: choosing Claude or Codex as your assistant reached Business plans on the same schedule as Pro, back in February 2026, well before this August's four-tier restructuring. That rollout is reported consistently enough across coverage of GitHub's changelog to treat as settled, even though the changelog page itself, on github.blog, wasn't reachable to quote directly while researching this article. What's newer, and what this article is actually about, is Pro+ and Max specifically unlocking delegation on top of that older assistant-selection access.

What "premium models" actually means per tier

GitHub's plans page doesn't publish one single master list of every model and which tier unlocks it, but the tiers do name specific models at specific points, and the pattern is consistent: Free names Haiku 4.5 and GPT-5 mini, smaller and cheaper models suited to a completions-capped tier. Pro+ is the first tier to explicitly name Opus. Nothing on the page states a complete Pro-tier-only model list separate from Free's named models and Pro+'s addition of Opus, so if a specific model's availability matters for your decision, the honest answer is to check the live plans page for that model by name rather than assume from tier level alone. What every tier's credit allotment actually draws down against, beyond the general "premium usage" framing, also isn't broken out on the page in enough detail to state precisely here.

Max ($100/month): who actually needs it

Max includes everything in Pro+, plus priority access to new models and features, roughly 2.9x more included usage than Pro+, and $200 of monthly credits. There's no new capability listed on Max that Pro+ doesn't already have, the difference is entirely headroom: more usage before you touch overage billing, and earlier access when GitHub rolls out something new.

That makes Max a volume decision, not a features decision. It's the right tier if you're running enough agent mode sessions, delegated background tasks, or premium-model requests that Pro+'s $70 credit allotment runs out most months, not because Max unlocks something Pro+ can't do at all.

Practically, that means the right way to size Max isn't reading its feature list, it's watching your Pro+ usage. If you're an individual developer who occasionally delegates a task to Claude Code and mostly writes code interactively, $70 of monthly credits is unlikely to be the constraint. If delegation is your default way of working, several background agents running through tickets across a working day, Pro+'s allotment is the thing that runs out first, and Max buys headroom rather than anything new to point at.

Business and Enterprise: the org-level tiers

Business and Enterprise sit outside the four consumer tiers entirely. GitHub's plans page describes them only as offering licence management, policy management and IP indemnity, without a fixed public per-seat price the way Free through Max are listed. If you need centralised billing, seat management across an organisation, or contractual indemnity coverage, that's a sales conversation rather than a self-serve upgrade, and it isn't simply "Pro+ but for teams."

For a team standardising on agent skills across repositories, that distinction matters more than it first looks. A .github/skills/ folder committed to a repo works identically regardless of which tier the developers reading it are on, so Business's actual value for a skills-heavy team isn't better skill support, it's the licence and policy controls that let an admin manage seats and access centrally rather than every developer picking their own individual tier.

Which tier do you need if you're using agent skills

Boiled down to the actual decision:

• Just want Agent Skills working with Copilot's own assistant? Free is enough. Nothing about .github/skills/ or agent mode requires a paid tier.
• Hit the Free completion cap, or want to pick Claude or Codex as your interactive assistant? Pro, at $10/month, covers both.
• Want to hand a task to Claude Code or Codex as a background agent and have it come back with a pull request? That's Pro+, at $39/month, the delegation feature isn't on Pro regardless of completion volume.
• Running enough of the above that Pro+'s credits run out most months? Max, at $100/month, is more headroom on the same feature set, not a new tier of capability.
• Need organisation-wide licensing, policy control or indemnity? That's Business or Enterprise, priced separately from the individual tiers.

A change worth flagging, and what's confirmed versus reported

Trade coverage from around June 2026 describes GitHub adding the Max tier and shifting Pro and Pro+ from a premium-requests model to usage-based GitHub AI credits at the same time. That would explain why current pricing talks in credits rather than the request-count-and-multiplier system Copilot used previously, where different models consumed a request budget at different multiples. GitHub's own blog announcement of that specific change lives on github.blog, which this article's research couldn't reach directly to confirm the exact date or the prior request-multiplier figures. Everything stated above about current pricing, features and tier boundaries is checked directly against GitHub's own live plans page rather than against that older billing history, so treat the history as reported context, not a verified fact this article is asserting on its own authority.

Troubleshooting

You're on Pro and Copilot won't delegate a task to Claude Code. That's expected, not a bug. Delegation to a third-party agent as a background task is a Pro+ and Max feature in preview. You can still select Claude interactively on Pro, but the assign-it-and-walk-away workflow needs the higher tier.

A skill in .github/skills/ isn't loading, and you're on Free. Tier isn't the cause. Check the folder is actually committed and not excluded by .gitignore, and that the SKILL.md frontmatter has a name and description Copilot's agent mode would recognise as relevant to what you asked, the same checklist that applies on any tier.

You upgraded to Max expecting a new feature and can't find one. There isn't one to find. Max's entire difference from Pro+ is usage headroom and priority rollout timing, not an additional capability gated behind it specifically.

Where to go next

For the mechanics of getting a skill running in Copilot's agent mode in the first place, see how to install agent skills in GitHub Copilot. For what Agent Skills are and how the format works across agents generally, see what are agent skills and the SKILL.md format explained. Browse the current GitHub Copilot catalogue at getclaudeskills.com/platforms/github-copilot, or everything cataloged at getclaudeskills.com/skills.