GitHub Copilot Switches to AI Credits Billing in June: Will You Pay More?

GitHub Copilot is switching from Premium Request Units (PRUs) to AI Credits token-based billing on June 1, 2026. Plan prices stay the same, but the billing logic changes completely — and whether you end up paying more depends heavily on which models you use and how agentically you work. Let’s break it down.

---

The Change in One Table

Old Model (until May 31)	New Model (from June 1)
Pro: 300 premium requests/month, overage $0.04/req	Pro: $10 in monthly AI Credits, overage billed per token
Pro+: 1,500 premium requests/month	Pro+: $39 in monthly AI Credits
Each “request” = one unit, regardless of token count	Charged by actual tokens × model rate

Code completions and Next Edit Suggestions remain free across all plans — they never consume AI Credits.

---

Monthly AI Credits by Plan

Plan	Monthly Price	Included AI Credits	USD Value
Copilot Free	$0	Limited	—
Copilot Pro	$10	1,000 Credits	$10
Copilot Pro+	$39	3,900 Credits	$39
Copilot Business	$19/user	1,900 Credits + 3,000 promo (Jun–Aug)	$19
Copilot Enterprise	$39/user	3,900 Credits + 7,000 promo (Jun–Aug)	$39

Conversion: 1 AI Credit = $0.01 USD. Once your monthly allotment runs out, you can set an overflow budget or simply stop until the next billing cycle.

---

What Consumes Credits vs. What’s Free

Always free (no credits consumed):

• Inline code completions
• Next Edit Suggestions

Consumes AI Credits:

• Copilot Chat
• Copilot Agent / Autopilot sessions
• Copilot CLI
• Copilot Cloud Agent
• Copilot Spaces & GitHub Spark
• Third-party Copilot Extensions
• Code Review (also consumes GitHub Actions minutes)

---

Model Token Pricing Table

Prices per 1 million tokens, converted to AI Credits (100 Credits = $1)

Model	Input (credits/M)	Cached Input (credits/M)	Output (credits/M)
GPT-5 mini	25	2.5	200
Raptor mini	25	2.5	200
Grok Code Fast 1 (xAI)	20	2	150
Gemini 3 Flash	50	5	300
GPT-4.1	200	50	800
GPT-5.2	175	17.5	1,400
Gemini 2.5 Pro	125	12.5	1,000
Claude Haiku 4.5	100	10	500
GPT-5.4	250	25	1,500
Claude Sonnet 4.x	300	30	1,500
GPT-5.5	500	50	3,000
Claude Opus 4.x (Pro+ only)	500	50	2,500

---

What Does a Typical Session Actually Cost?

Scenario A: Regular Chat

• Input: ~2,000 tokens (question + code context)
• Output: ~500 tokens (response)

Model	Cost per chat (credits)	Sessions on Pro’s 1,000-credit budget
GPT-5 mini	0.15	~6,600
Claude Haiku 4.5	0.45	~2,200
Claude Sonnet 4.x	1.35	~740
Claude Opus 4.x	2.25	~444 (Pro+ only)

Scenario B: Agentic Coding Session

Agents read multiple files, so context balloons significantly:

• Input: ~15,000 tokens
• Output: ~3,000 tokens

Model	Cost per agent task (credits)	Tasks on monthly budget
GPT-5 mini	0.98	~1,020 (Pro)
Claude Haiku 4.5	3.0	~333 (Pro)
Claude Sonnet 4.x	9.0	~111 (Pro)
Claude Opus 4.x	15.0	~260 (Pro+, $39 budget)

---

More Expensive or Cheaper Than Before?

It depends entirely on your model choice and session size:

Likely cheaper (light users):

• Primarily using GPT-5 mini or Grok Code Fast for daily chat
• Under the old model, 300 premium requests was a hard cap — the new model gives equivalent usage with credits left over

Likely more expensive (heavy agentic users):

• Running frequent Claude Sonnet or GPT-5.4 Agent sessions on large codebases
• Under the old model, 300 requests was a fixed ceiling. Under the new model, 111 Sonnet agent sessions already exhaust Pro’s entire $10 credit budget
• Bottom line: longer context + pricier models = faster credit burn

Community reaction: A vocal segment of developers complained the change means “you will get less, but pay the same price” — particularly those relying on long-context sessions in large repos. GitHub’s official position is that Copilot has evolved from a chat assistant into an agentic platform, and token-based billing better reflects the actual compute involved.

---

How to Control Your Spending

The new billing model introduces budget controls at three levels:

1. Individual users: Set a monthly overflow cap in billing settings. Setting it to $0 stops usage entirely once credits are exhausted — no surprise charges.
2. Organization / cost center: Admins can set per-user budgets across teams.
3. Enterprise level: Centralized control across all seats.

Practical tips:

• Unsure about your usage? Set overflow cap to $0 for the first month and monitor actual consumption before enabling overages.
• Heavy agent user? Estimate: monthly agent sessions × credits per session = expected spend. Budget accordingly.
• Save credits: Swapping from Claude Sonnet to Gemini 3 Flash or GPT-5 mini for routine tasks cuts costs by 85%+ with minimal quality impact for simple queries.

---

Migration Timeline

User Type	Migration Date
Monthly Pro / Pro+	Auto-migrates June 1, 2026
Annual Pro / Pro+	Stays on current plan until annual renewal
Business / Enterprise	Switches June 1 (with promotional credits Jun–Aug)

---

Recommendations by User Profile

Profile	What to do
Light user (a few chats/day, cheap models)	Minimal impact — credits will be ample
Moderate user (daily chat + occasional agents)	Pro $10 should suffice; set overflow cap to $0–$5
Heavy agentic user (Sonnet/Opus + large context)	Consider Pro+, or evaluate Cursor / Cline + direct API
JetBrains / deep GitHub integration user	Copilot remains the best fit; monitor monthly usage

If the new billing pushes your costs up significantly, see our Claude Code budget alternatives guide or how to cut Cursor costs for alternative setups.

---

Sources: GitHub Copilot Official Blog · GitHub Docs: Models and Pricing · Verified: 2026-05-11