GitHub Cut Copilot Credits 37% and Claude's API Got 50% More Expensive, Same Week
On September 1, GitHub cut the Copilot credits included in Business and Enterprise plans by roughly 37% and 44% respectively, from 3,000 to 1,900 per user per month on Business and 7,000 to 3,900 on Enterprise, while the sticker price stayed put at $19 and up. The same week, Anthropic's introductory Claude Sonnet 5 API rate expired, moving pricing from $2 input / $10 output per million tokens to $3/$15, a straight 50% increase on both legs. Neither change showed up as a price increase on a pricing page. Both showed up as a smaller number of credits or a bigger bill at the end of the month.
The specific numbers
GitHub Copilot Business dropped from 3,000 to 1,900 included credits per user per month. Enterprise dropped from 7,000 to 3,900. The plan price didn't move: Business is still $19 and up per seat. Claude Sonnet 5's API pricing moved from $2/$10 per million input/output tokens to $3/$15, a clean 50% jump on both sides, once the introductory rate lapsed. Two different vendors, two different mechanisms, the same underlying effect: whatever you were paying for a given amount of usage in August costs more in September, even though nothing on the plan page changed.
What didn't change, and why that matters
Claude's $20 Pro subscription and the $100/$200 Max tiers were unaffected, the pricing change only hit the metered API. Cursor held its Hobby (free) and Pro ($20/month) tiers steady through the same period. So the actual impact on a given solo builder depends entirely on which product tier they're using: someone paying a flat Claude Pro subscription for personal coding assistance saw no change at all, while someone calling the Claude API directly from a script or an agent pipeline took the full 50% hit on every token processed.
Why a credit cut is the more dangerous version of a price increase
A straightforward price hike is visible: the number on the plan page goes up, you notice, you decide whether to keep paying it. A credit cut behind an unchanged sticker price is designed not to trigger that moment. You keep seeing "$19/month" and assume nothing changed, until you either hit the new, lower ceiling mid-month or the overage charges start showing up on an invoice you don't read closely. For a solo operator running lean, that's the worse failure mode, because it doesn't prompt a decision, it just quietly changes the unit economics of whatever you built assuming the old allowance.
What to actually check this week
The concrete move isn't reading the pricing page, it's checking the usage meter. If you're on GitHub Copilot Business or Enterprise, look at last month's actual credit consumption against the new, lower allowance, not the old one you budgeted around. If you're calling the Claude API directly, whether through a script, an agent framework, or a content pipeline like the one that writes this blog, check whether last month's token volume at the new $3/$15 rate still fits your margin, especially if you're running the API inside a product you charge customers for. If you're on a flat subscription tier for either vendor, this specific change doesn't touch you, which is itself worth confirming rather than assuming.
The honest take
I budget my monthly tool stack assuming prices only go up over time, but I'd been checking the wrong number. The plan name and the sticker price are the last things a vendor changes when they want to raise revenue without triggering a churn conversation, the credit allowance or the per-token rate moves first and quietly. My actual rule going forward: every time a vendor announces anything, even something that sounds unrelated to pricing, I check the usage meter on my own account before assuming last month's bill will look like this month's.
Where I could be wrong: both of these are early-in-the-cycle changes, and it's possible competitive pressure pushes one or both vendors to reverse or soften them once the churn data comes in. Cursor holding its Pro tier steady through the same window is itself a signal that at least one competitor sees an opportunity in not moving right now. I wouldn't assume this specific credit allowance or this specific per-token rate is permanent, but I would assume the pattern (meter moves before sticker price does) keeps repeating.
Author
Lukas
@lukcombinator