Blog · Industry
Industry

What GitHub Copilot Costs After June 1

CU
Collier Uesseler
May 13, 2026·6 min read

On May 12, 2026, GitHub announced that individual Copilot plans are moving to usage-based billing starting June 1. The flat premium-request allotment system goes away. In its place: AI credits, denominated in dollars, priced against the model you actually use. If you have a team on Copilot Pro or Pro+, your June bill will be the first one that varies by behavior rather than by seat count. Most teams have no idea what that number will be.

What changes on June 1

Today, paid Copilot plans include a fixed allotment of premium requests per month. Use fewer, pay the same. Use more, get cut off or pay per request. It's a blunt instrument, but it's predictable.

Starting June 1, that changes to an AI credits system. Each plan comes with a base credit pool and a "flex allotment" on top. The credits are denominated in dollars and consumed at rates that vary by model. The published plan structure is:

$15
Copilot Pro ($10/mo) — $10 base + $5 flex included before overages apply
$70
Copilot Pro+ ($39/mo) — $39 base + $31 flex included
$200
Copilot Max ($100/mo) — $100 base + $100 flex included

Usage beyond the included credits bills at the model's per-token rate. Code completions and next-edit suggestions remain unlimited on all paid plans. Everything else — chat, agent mode, Workspace queries, code review — draws from the credit pool.

Note on per-model token rates: GitHub's docs page listing exact per-model $/token rates was not publicly accessible at publication time. The examples below use the AI credit pool sizes as the unit of measure rather than specific per-token figures. Before June 1, verify your plan's model rates in Settings → Billing → Usage on github.com, where GitHub is publishing April usage previews in AI credit terms.

The math, by usage profile

These are estimates for three hypothetical usage patterns. They are not customer data. Use them as a calibration tool, not a budget number.

Profile A Occasional chat user — Copilot Pro
Pattern: ~5 short chat exchanges per workday. Asking questions about code, quick explanations. Rarely uses agent mode. Mostly GPT-5 mini or the default model.

Workdays/month: 22
Sessions: ~110 chat turns/month, short context, minimal tool calls
Completion usage: Unlimited, not counted
Well within $15
included credits
This profile is unlikely to see overages. The risk here is not cost — it's that the bill becomes unpredictable the moment one developer on the team starts running agents.
Profile B Daily agent user — Copilot Pro+
Pattern: Uses agent mode daily. Typically 3-5 agent sessions per day, each involving 8-15 turns with file reads, edits, and tool calls. Uses Claude Sonnet or GPT-4o for most sessions. Longer context windows from open files.

Workdays/month: 22
Estimated sessions: ~80 agent sessions/month
Context per session: Several thousand tokens in, hundreds out per turn
$40–$80
estimated monthly credits
Estimate for this profile only. At the low end, stays within the $70 Pro+ pool. At the high end, hits overages. Actual cost depends on model selection — verify your model's rate in Settings before June 1.
Profile C Power user — large repo, continuous agent sessions
Pattern: Copilot Workspace for multi-file tasks. Runs extended agent sessions on large codebases, sometimes for 30-60 minutes on a single task. Uses the highest-capability models available. 5-8 substantial sessions per day.

Workdays/month: 22
Estimated sessions: ~120-160 extended agent sessions/month
Context per session: Large — many open files, long conversation history
$100–$200+
estimated monthly credits
This profile exceeds Pro+ included credits ($70) under most scenarios. Max ($200 included) may cover moderate months. Heavy months will see overages regardless of plan. Community reports of $30-40 per single agentic session with top-tier models are consistent with this range.

The wide ranges above are not imprecision on my part. They reflect genuine uncertainty in the model: GitHub has not published a simple "this model costs X per query" table that an end user can bookmark. Your actual bill depends on which model each developer defaults to, how long their context windows grow, and whether anyone on the team is running the kind of deep agent sessions that the platform actively encourages.

Why this is hard to forecast without help

GitHub's Settings page shows you Copilot usage for your account. What it does not show:

  • Aggregate across your team. If you have 8 developers on Copilot Pro, you need to check 8 accounts, or wait for the organization-level report. There is no "total AI credit burn for my org this month" in a single view.
  • How Copilot spend compares to Cursor. If half your team uses Cursor and half uses Copilot, there is no place in either product's dashboard that shows you both. You are triangulating from two separate billing pages.
  • How your subscription spend compares to your API spend. If you also have a direct Anthropic or OpenAI account, that is a third billing page with a different unit system entirely.
  • What you were on track to spend before you hit the bill. The April usage reports GitHub released for preview are acknowledged by GitHub itself to contain gaps and duplicate entries. They are directional, not reliable for budget planning.
"The usage reports, budgeting — all of it ... confusing and nonsensical. The Billing and Billing History pages are ..." GitHub community discussion #192948 — active thread, 100+ replies, May 2026

That thread has developers asking how to forecast June before they get surprised by a bill. The honest answer right now is: you cannot do it precisely with the tools GitHub provides. You can get a rough sense by downloading the April preview report and applying some arithmetic, but you cannot watch it in real time, and you certainly cannot watch it alongside your other AI spend.

What Stax does about this

Stax has a native GitHub Copilot integration. You connect your account once (see the connection guide), and Copilot spend lands in the same dashboard as your Cursor subscription, your Anthropic API bill, your OpenAI costs, and anything else you have connected. One number, updated on a schedule, with a spend alert if you set one.

That is the part Helicone never built and that GitHub's own dashboard cannot show you: total AI spend across vendors, not per-vendor spend across tabs. If you have 10 people on Copilot Pro+ and 5 on Cursor Business, Stax shows you one line per vendor and one total at the top. You can set a threshold — say, $3,000/mo across all AI tools — and get an email before you blow through it.

Stax is $29/mo for founding members (flat, not per seat) during early access. Join the waitlist at stax.report/signup.

Copilot's billing is about to get variable. Your dashboard for all of it should not be five tabs and a spreadsheet.

Join at $29/mo →
More from Stax

Worth reading next.

All posts →
Founder update May 13, 2026

Stax Is Ready. Here Is What We Shipped.
Founding post May 11, 2026

Why I built Stax: 4 weeks reading Reddit, one product.
Comparison live now

Stax vs. Helicone — what to do if you need to migrate.
Security live now

How we handle your credentials, in plain English.