GitHub Copilot moves to token-based billing — flat-rate is dead, power users seeing 10-50× bill spikes
TL;DR: GitHub Copilot transitioned all plans to usage-based billing on June 1, 2026 — today. Base subscription prices unchanged: Copilot Pro $10/month, Pro+ $39/month, Business $19/user/month, Enterprise $39/user/month. What’s different: each plan now includes a monthly allotment of GitHub AI Credits in an amount equal to the subscription dollar ($10 plan = $10 in credits, $39 plan = $39 in credits, etc.). Tokens consumed by every input, output, and cached-token request — across Copilot Chat, Agent Mode, Workspace, and any premium-model request — count against that allotment, billed at the listed API rates per model. What stays free: code completions and Next Edit suggestions remain unmetered. What’s hurting: power users running agentic workflows on premium models report 10-50× bill spikes versus the old flat-rate model. The “unlimited Copilot for $10” era is over — and the AI Harnesses category just became materially more expensive for the enterprise-default option.
What changed today
The reporting from GitHub’s official blog, GitHub docs, Tech Times, how2shout, Windows Forum, and Enterprise DNA confirms the structural shift effective June 1, 2026:
The pricing structure
| Plan | Monthly fee | Monthly AI Credits | Net effective free usage |
|---|---|---|---|
| Copilot Pro | $10 | $10 in credits | $10 of token consumption |
| Copilot Pro+ | $39 | $39 in credits | $39 of token consumption |
| Copilot Business | $19/user | $19/user in credits | $19/user of token consumption |
| Copilot Enterprise | $39/user | $39/user in credits | $39/user of token consumption |
Once a user exceeds the included credits, paid plans can purchase additional AI Credits at the published API rates per model. The credit-to-token conversion is the standard published API rate for each model (e.g., Claude Sonnet 4.7 at $3 input / $15 output per million tokens, GPT-5.5 at corresponding rates).
What still works without metering
- Code completions (the original inline-suggestion product) — free, unmetered
- Next Edit suggestions — free, unmetered
Both of these were the original Copilot value proposition. They remain available to all subscribers at any tier without consuming credits.
What now consumes credits
- Copilot Chat (the conversational interface)
- Copilot Agent Mode (autonomous multi-step work)
- Copilot Workspace (the planning + execution surface)
- Any premium-model request (Opus, GPT-5.5, etc.)
- Any tool-calling or RAG-augmented interaction
In effect: the modern “AI-as-a-coworker” Copilot features all bill by token now. The legacy “AI as tab-complete” features remain free.
Why this matters structurally
The flat-rate era for AI coding tools is functionally ending. GitHub Copilot was the last major harness still operating on a “pay subscription, use unlimited within fair-use” pricing. Claude Code has hard rate limits but no per-token billing inside them. Cursor operates on similar subscription-plus-soft-limit logic. OpenAI Codex bills hours of work via ChatGPT Plus. With Copilot moving to explicit token-based metering, the entire AI Harnesses category is now structurally on a credits or tokens or hours economy — flat-rate predictability is gone.
Power-user economics shifted overnight. Reports from the developer community describe agentic workflows that previously fit comfortably inside the $10 Pro subscription now generating monthly bills of $100-$500+. The Tech Times report cites 10-50× bill increases for the heaviest Copilot Agent users. This isn’t a pricing increase per se — the underlying API rates are unchanged — but rather a removal of the cross-subsidy that made flat-rate Copilot cheap for power users.
The enterprise procurement value proposition narrows. GitHub Copilot’s structural win in the AI Harnesses category was procurement convenience: “if your company already pays GitHub Enterprise, Copilot is an upsell, not a vendor evaluation.” That’s still true. But the cost-predictability story — “you know exactly what each developer will cost per month” — is broken. Procurement teams will now have to forecast token consumption per seat, which is hard, especially in early agentic-workflow rollouts where usage spikes are common.
How it compares to the alternatives
For a power user running ~$50/month equivalent of token consumption:
| Tool | Effective monthly cost | Predictability |
|---|---|---|
| Claude Code | $20 (Pro) → $100 (Max 5×) → $200 (Max 20×) | High — rate limits, no per-token bill |
| Cursor | $20 (Pro) → $40 (Business) | High — soft-limited subscription |
| OpenAI Codex | $20 (ChatGPT Plus) → $200 (Pro) | Medium — included with ChatGPT tiers |
| GitHub Copilot (new) | $10 base + $40 in overage credits = ~$50+ | Low — variable token consumption |
| OpenCode (BYO API) | $20-$100 of API spend | Low (but transparent) |
For the same workload, GitHub Copilot’s old flat-rate $10 plan was the cheapest entry. After today, it’s competitive at the low-usage end but loses the price advantage entirely for power users — Claude Code Max 5× ($100) or Cursor Business ($40) become better deals for any user generating $40+ of token consumption.
What it means for GitHub Copilot users
Light users (mostly tab-complete): nothing changes. Code completions stay free; you’ll likely never touch your $10 credit allotment.
Moderate users (occasional Chat / Agent use): monitor the credit dashboard for the first month. If you’re consistently consuming under $10, the new model is functionally equivalent. If you’re exceeding $10, evaluate whether bumping to Pro+ ($39 + $39 credits = $39 effective monthly cap before overage) makes sense.
Power users (heavy Agent Mode / Workspace): this is the group most affected. The flat-rate cross-subsidy that made Copilot the cheapest serious AI coding tool is gone. Evaluate Claude Code Max 5× ($100/month for substantially more headroom + best-in-class capability) or Cursor Business ($40/month, multi-model flexibility).
Enterprise teams: procurement workflow needs adjustment. Per-seat cost forecasting now requires assumptions about token consumption per developer per month. Expect early rollouts to produce surprise bills as teams ramp up agentic usage. Setting up team-level usage caps and per-developer credit limits in the GitHub admin console will be the operationally critical step.
What it means for the broader AI Harnesses category
This pricing shift accelerates a structural trend: harness pricing is consolidating around either rate-limited subscriptions (Claude Code, Cursor) or pay-per-token billing (GitHub Copilot now, OpenCode BYO-key always). The flat-rate-with-fair-use model is functionally dead in the category.
For tool readers, the implication is that “AI coding tool pricing” is no longer a static number — it’s a function of how you use it. The right shopping question shifts from “what does it cost?” to “what does it cost for my workload?”
For the AI Harnesses roundup, the GitHub Copilot recommendation framing tightens: “wins on procurement convenience for GitHub-Enterprise shops” still holds, but the “wins on cost predictability” claim has to come off the table.
The honest caveats
Two caveats worth surfacing:
Most users won’t notice. Approximately 80% of Copilot subscribers primarily use code completion and Next Edit — both of which remain free. The pricing change is structurally significant for the agentic-workflow power-user cohort, but most subscribers will see their bills unchanged.
The new pricing is more transparent, not necessarily worse. Token-based billing has the merit of making AI compute costs visible. Power users who were previously cross-subsidized are now paying their actual cost. From a market-design perspective, this is more honest than the flat-rate model that was always going to be unsustainable as agentic workflows scaled.
What it changes for Pick Right readers tomorrow
If you’re a GitHub Copilot subscriber, check your GitHub billing dashboard in the first week of June to see your actual credit consumption. If you’re a light user, ignore this story. If you’re a power user, evaluate whether Claude Code at $20-$100/month or Cursor at $20-$40/month is now structurally cheaper for your usage pattern.
For broader context, see the GitHub Copilot review, the AI Harnesses category roundup, the Claude Code review, the Cursor review, and the Claude Code vs Antigravity comparison for the broader harness-pricing picture.
Sources
Related tool reviews
Questions or corrections? Email Pick Right. Want the full list? See all news.