GitHub Copilot Switches to Token-Metered Billing: What It Signals for SaaS Pricing

What happened

On June 1, 2026, GitHub moved every Copilot plan to usage-based billing, retiring the Premium Request Units that had governed paid usage and replacing them with token-metered "GitHub AI Credits." The new unit is deliberately simple: 1 AI Credit equals $0.01, and consumption is calculated from token counts, input, output, and cached, charged at the published API rate for whichever model a developer selects. Instead of an abstract "request" that could mean almost anything, billing now tracks the raw compute a session actually burns.

Each plan now ships with a monthly credit allotment baked into its existing price. Copilot Pro continues at its prior price point with $15 in included credits (1,500 credits), Pro+ includes roughly $70 (7,000 credits), and a new Max tier at $100 per month carries $200 in credits (20,000). On the team side, Copilot Business stays at $19 per user with $19 in credits, and Enterprise stays at $39 per user with $39 in credits, with pooled monthly allowances of 1,900 and 3,900 credits per seat respectively. GitHub layered on temporary promotional allowances of 3,000 and 7,000 credits per seat from June 1 to September 1, 2026, softening the transition for organizations.

Crucially, the everyday surface most developers touch stayed free. Code completions and Next Edit suggestions remain unlimited on paid plans and consume no credits at all. What meters is the heavier, agentic work: chat, multi-file edits, and model-intensive reasoning. Model choice now drives cost directly, GPT-5.5 is listed around $5 per million input tokens and $30 per million output, while a lighter model like MAI-Code-1-Flash runs roughly $0.75 and $4.50, a spread that can mean a 24x cost difference for the same task depending on which engine a developer picks.

Why it matters for practitioners

For founders shipping AI features, Copilot is the most-watched pricing experiment in software right now. A category-defining product owned by Microsoft just abandoned a flat, predictable "requests" model for raw token metering, and the way it did it is a template worth studying closely.

1. Separate the cheap habit-forming surface from the expensive value surface. GitHub kept completions free and unlimited while metering the agentic work. That is a textbook structure: the always-on feature that creates the habit stays free, and the high-compute feature that delivers outsized value carries the meter. If you sell AI, your equivalent of "completions" is whatever keeps users in the product daily, and your equivalent of "agent runs" is what you charge for. Copilot's design sits squarely in the developer-tools landscape where this split, free editor assistance, paid heavy lifting, is becoming the default.

2. Your included allotment is a free tier in disguise. Bundling $15 or $19 of credits into a paid plan is functionally a metered free allowance with a floor under it. The same discipline that governs a good free tier, sizing the included amount so most users stay inside it while heavy users predictably spill over, now governs credit allotments. Set the allowance too low and you create "meter shock" that drives churn; set it too high and you give away the expansion revenue the model exists to capture.

3. Token metering exposes your model-cost pass-through. By pricing each model at API rates, GitHub made its own cost structure visible to customers and tied its margin to model selection. For a bootstrapped founder, that is the double-edged reality of token-based pricing: it protects you from runaway COGS on power users, but it also means a single expensive model can blow up a customer's bill, and your support queue, overnight.

Key details

• Effective date: June 1, 2026, across all Copilot plans
• New unit: GitHub AI Credits, 1 credit = $0.01 USD
• Metering basis: input, output, and cached tokens at each model's listed API rate
• Individual credits: Pro ~$15 (1,500), Pro+ ~$70 (7,000), Max ~$200 (20,000) at $100/mo
• Team credits: Business $19/seat (1,900 pooled), Enterprise $39/seat (3,900 pooled)
• Promotional credits: 3,000 (Business) and 7,000 (Enterprise) per seat, June 1–Sept 1, 2026
• Stays free: code completions and Next Edit suggestions, unlimited on paid plans
• Model spread: up to ~24x cost difference between the priciest and cheapest models
• PR review: Copilot pull-request review now counts against included Actions minutes

Market implications

The deeper signal is that even the largest vendors no longer trust a fixed unit to track AI value. Premium Request Units worked when the underlying compute per request was roughly stable. The moment models ranged from cheap-and-fast to expensive-and-deep, any flat unit either overcharged light users or bled margin on heavy ones. Token metering is GitHub's admission that the only honest unit for AI is the compute itself, and that buyers will increasingly be asked to internalize the cost of their model choices.

For independent founders, this is more opportunity than threat. Microsoft's scale forces it toward complex, exposed pricing that invites "meter shock" backlash and a cottage industry of cost-optimization guides. A smaller, product-led company can do something Copilot structurally cannot: keep pricing legible. You can choose one or two models, bundle a generous allowance, and present a single predictable number while still letting usage expand revenue underneath. Pricing as a growth lever rewards clarity, and clarity is exactly what gets harder at Copilot's scale.

The practical move for 2026 is to treat Copilot's redesign as a live case study rather than a mandate. Token metering is the right instinct when model costs vary wildly and power users would otherwise erode your margin. But the parts worth copying are the structure, free habit surface, metered value surface, an allowance sized like a free tier, not the complexity. The founders who win on AI pricing will capture the same expansion GitHub is chasing while sparing customers the spreadsheet math that now comes standard with a Copilot subscription.

Related resources

• Developer Tools Analysis, Where Copilot and AI coding assistants sit in the independent developer-tool market
• How to Launch a Free Tier, Sizing included allotments so most users stay free and heavy users expand
• Product-Led Growth Glossary, Why pricing structure is one of the strongest PLG growth levers