How to predict when your AI tool will nerf you
Sit with the gotnerfed index for an afternoon and a pattern shows up. Pricing nerfs aren’t random. They’re lagging indicators of pressure that’s been building for weeks - sometimes months. If you know what to watch, you can usually call the change before the email lands in your inbox.
Here are the six signals that almost always show up first, scored by how reliable each one has been in the receipts we’ve catalogued since 2024.
1. The free tier suddenly “upgrades”
When a vendor announces that the free tier is getting “more powerful” or moving to a “better model,” the actual change is almost always a reduction in volume. The marketing is the leading indicator. Watch for the words “upgrade,” “improved,” or “modernized” near anything that mentions limits, credits, or daily caps.
Confidence: high. We’ve seen this pattern preceded eight of the last ten free-tier nerfs in our database.
2. A new enterprise tier appears
Vendors don’t introduce a $500/mo tier in isolation. A new top tier almost always signals a coming squeeze on the tier below it - either price hikes, feature removals, or quota cuts - to push users up the ladder. The interval is usually 30 to 90 days between “Enterprise launched” and the next pricing change on Pro/Team.
3. The model name silently changes
“Claude 3.5 Sonnet” becomes “Claude latest.” “GPT-4” becomes “GPT” with no number. When a vendor strips a version number from a product surface, they’re reserving the right to swap the model underneath without notifying you. This is the #1 most common “quality nerf” pattern we track.
4. A funding round closes
Counterintuitive but consistent: companies raise money, then six to twelve months later they raise prices. The capital lets them grow into a bigger price point, and the board pressure to show revenue acceleration forces it. Track Crunchbase or just watch for “Series” in tech press - mark the date and check back in 180 days.
5. Compute provider news
When NVIDIA, AWS, or Azure raises GPU pricing, the cost passes downstream within a quarter. If you see “H100 supply” or “inference cost” in an earnings call from your provider’s provider, your provider’s pricing is about to move.
6. The status page suddenly gets a lot of yellow
Frequent degradations on a public status page are a sign of overcapacity. Vendors don’t solve overcapacity by buying more compute - that’s expensive. They solve it by reducing demand. That means caps, throttling, or a quietly cheaper model in the request path.
What to do when you spot one
- Open a parallel account on a competitor that lets you import context. Cursor → Windsurf, ChatGPT → Claude, v0 → Lovable. Don’t wait until you’re locked out to learn the migration path.
- Export everything you can - conversations, prompts, custom instructions. The day a tier disappears is the worst day to learn that the export button is gated to the tier you no longer have.
- Cancel the annual plan if you’re close to renewal. Annual locks you into pre-nerf pricing on the surface, but most contracts have a “we can change this at any time” clause. You’re paying for protection that doesn’t exist.
- Subscribe to the watchdog (us). The whole point of gotnerfed is that you don’t have to watch for these signals manually - we do it across every AI tool we track and send you the receipt when one fires.
The trend is the product
AI inference is still subsidized. The capital flowing into model providers, IDEs, and agent tools is propping up unit economics that won’t survive a normalization. Every silent nerf you see is a small step toward the price the market actually pays. The question isn’t whether your tool will nerf you. It’s when, and whether you’ll have an alternative ready.
That’s what this site is for.