<h2>What’s actually changing at GitHub Copilot</h2>
<p>Starting June 1, 2026, GitHub is moving all Copilot plans to usage-based billing. The flat-rate “premium requests” model is out. In its place: GitHub AI Credits, calculated on token consumption rather than raw message counts.</p>
<p>Base prices stay the same — Pro at $10/month, Pro+ at $39/month, Business at $19/user/month, Enterprise at $39/user/month. Each plan includes a monthly AI Credits allotment. Code completions and Next Edit suggestions remain free and don’t touch your credits. Business and Enterprise customers get extra credits through August to ease the transition, and a billing preview launches in early May so users can see projected costs before the switch hits.</p>
<p>Admins get new budget controls at the enterprise, cost center, and individual user level, plus pooled usage across organizations.</p>
<h2>Why this is happening now</h2>
<p>The short answer: agentic workflows broke the math.</p>
<p>Copilot’s original pricing was designed for autocomplete — lightweight, fast, cheap per interaction. Agentic coding sessions are something else entirely. They run in parallel, they run long, and they consume compute at a completely different scale. One developer on a $40/month plan sent a single message that burned through 60 million tokens and roughly $30 of inference — and was still going. Fifteen messages later, the same user had consumed $221 worth of tokens while barely touching 2% of their plan’s nominal limit.</p>
<p>GitHub’s weekly infrastructure costs reportedly doubled since the start of 2025. The economics of flat-rate billing simply don’t hold when a single session can cost more than a user’s entire monthly fee.</p>
<h2>The immediate fallout: paused sign-ups and tighter limits</h2>
<p>Before the June billing switch, GitHub already moved to contain the damage. On April 20, it paused new sign-ups for Copilot Pro, Pro+, and Student plans. Copilot Free is the only plan still accepting new users, with no timeline given for reopening the others.</p>
<p>Existing subscribers are seeing tighter limits too. Usage caps are now based on token consumption and model multipliers rather than request counts — which means you can exhaust your session limit while still having premium requests remaining, depending on which model you’re using. Anthropic’s Opus models have been removed from the Pro plan entirely; Opus 4.7 is now Pro+ only. GitHub is nudging users toward lower-multiplier models and away from parallelized workflows.</p>
<p>Usage visibility is being added to VS Code and Copilot CLI so users can actually see where they stand. Existing subscribers unhappy with the changes can cancel and request a refund for April usage by contacting support before May 20.</p>
<h2>GitHub isn’t alone</h2>
<p>Anthropichas been going through its own version of this. The $20/month Pro plan’s access to Claude Code was quietly removed from the pricing page, then reversed within hours after backlash on Reddit, Hacker News, and Twitter. An Anthropic employee described it as an A/B test affecting roughly 2% of new signups — but the updated pricing page was publicly visible and archived, and the reversal came only after the story spread. Separately, documentation suggested Opus model access in Claude Code would require additional payment beyond the Pro tier.</p>
<p>The $20 Pro plan was already strained before any of this. Five prompts into a Claude Code session and you’d hit usage limits. The plan was nominally available but practically unusable for serious agentic work.</p>
<p>OpenAI has made similar moves with Codex. Cursor has restructured its pricing. The pattern is consistent across the industry.</p>
<h2>The structural problem underneath all of this</h2>
<p>GitHub’s position is particularly awkward because it doesn’t own a foundational model. It’s reselling compute from OpenAI, Anthropic, and Google — each with their own markup. When those providers raise prices or tighten access, GitHub absorbs the hit. Cursor built its own inference layer (Composer) specifically to control this. GitHub didn’t, and is now paying for that decision.</p>
<p>IDC forecasts that Global 1,000 companies will underestimate AI infrastructure costs by 30% through 2027. That’s not a billing dashboard problem — it’s an architectural one. Better visibility into token consumption helps, but it doesn’t change the underlying cost structure.</p>
<p>The broader shift is real: AI coding tools are no longer fixed-cost productivity features. They’re metered infrastructure, and enterprise buyers should evaluate them that way — with usage ceilings, cost modeling, and architectural decisions about which models to use for which tasks.</p>
<p>Flat-rate AI subscriptions made sense when the workloads were predictable. Agentic workflows aren’t predictable. The subsidized era is ending, and the pricing is catching up.</p>


Collections

GitHub is overhauling Copilot billing starting June 1, 2026, replacing flat-rate premium requests with usage-based AI Credits tied to token consumption. Base plan prices remain unchanged, but the shift is driven by agentic coding workflows that can burn through months of compute in a single session — one user consumed $221 worth of tokens in 15 messages. Before the switch, GitHub has paused new sign-ups for Pro, Pro+, and Student plans, tightened usage caps, and removed Anthropic Opus models from the Pro tier. The piece contextualizes this within a broader industry pattern: Anthropic quietly restricted Claude Code access on its $20 plan, OpenAI restructured Codex pricing, and Cursor rebuilt its inference layer. GitHub's structural disadvantage is that it resells compute from OpenAI, Anthropic, and Google without owning a foundational model, leaving it exposed when provider costs rise. The conclusion: AI coding tools are now metered infrastructure, not fixed-cost subscriptions, and enterprise buyers need to plan accordingly.

GitHub pauses new Copilot sign-ups as agentic workflows break the pricing model

<p>No wonder why I was getting the weekly session limits reached with every new prompt XD</p>
