1,800 SaaS Pricing Changes in a Year: What the Top 500 Vendors Actually Learned

The top 500 B2B software and AI companies made more than 1,800 pricing and packaging changes in 2025. That is 3.6 changes each, close to one a quarter. Almost none of those SaaS pricing changes were structural. The year taught vendors something narrower and more useful: price is now edited the way product is edited, in small reversible increments, and the edits that fail are usually the ones nobody explained.

What "1,800 pricing changes" actually counts

The figure comes from the PricingSaaS 500 Index, which monitors the published pricing pages of 500 leading B2B software and AI companies. Rob Litterst, who co-founded the firm behind the index, put the 2025 total at more than 1,800 changes in a January 2026 write-up for Growth Unhinged.

It is the best public dataset on this question. It is also narrower than the headline suggests, in two specific ways that change how you should use it.

It only reads public pricing pages

The index observes what a vendor publishes. Negotiated enterprise agreements, private floor-price increases and quiet discount-policy changes produce no visible event.

That matters because the largest contracts in software are the least public ones. A vendor that raised its minimum commitment inside every new master agreement in 2025 registered zero changes here. The index measures published pricing behaviour, not pricing behaviour.

A "change" is not a fixed unit

Renaming a plan counts as one change. Rebuilding an entire product line onto a consumption meter also counts as one change. The average of 3.6 tells you how often a pricing page was edited, not how far the price moved.

This is why I read 1,800 as evidence of cadence rather than evidence of upheaval. High edit frequency with low structural change is exactly what you would expect from teams treating pricing as a product surface, which is the real shift.

One comparison to avoid. An earlier Growth Unhinged analysis by Litterst and John Kotowski counted 339 changes in 2024, split 126 pricing and 213 packaging. The company universe described in that piece is not the same 500-company index. Dividing 1,800 by 339 gives a growth rate that the underlying data does not support.

Lesson 1: packaging moved more than price

Break the 1,800 changes by type and the year has a clear shape. 43% were feature packaging updates, 40% were pricing structure changes and 17% were usage limit adjustments.

Packaging is the cheaper experiment. Moving a feature between tiers changes what a buyer receives without triggering the conversation a price rise triggers. If it lands badly you move it back, and almost nobody outside the affected cohort notices.

A price rise is the opposite. It is visible, it is discussed in renewal meetings, and reversing it is an admission. So the ratio in the data is not a curiosity. It is vendors routing around risk in a rational order.

The practical read for an operator is uncomfortable. If your team spent last year debating a price increase and never touched the packaging, you ran the expensive experiment and skipped the cheap one. Most of the learning in this dataset sits in the 43%.

Lesson 2: seats did not die, they absorbed the AI

The loud prediction going into 2025 was the end of per-seat pricing. Agents do the work, so why pay per human. The data did not cooperate.

Instead the seat became the delivery vehicle for AI. Notion, Slack and Loom bundled AI add-ons into their core plans and raised per-user prices by roughly $2.50 to $5 per user per month. Airtable and Atlassian went further and bundled AI features with credit limits at no added cost.

Per-seat pricing also held its position at the top of the market. In Kyle Poyar's May 2026 survey of 230 companies, 29% of companies above $150 million ARR still name per-seat as their primary model. Hybrid pricing leads overall at 37%, up from 25% a year earlier.

The uplift is small on purpose

An increase of $2.50 to $5 per user is small enough to clear a renewal without procurement escalation. It arrives as a better plan rather than a new bill. That is not timidity, it is the cheapest available route to a permanently higher base price.

If I ran a horizontal software product today I would take the small seat uplift over a separate AI SKU, and I would expect to be told that is unambitious. The separate SKU forces a fresh buying decision on every account. The uplift forces none. Seat economics are under real pressure from agents, the subject of the piece on seat compression and what replaces the headcount metric. But pressure on a model is not the same as its replacement.

Lesson 3: credits became the compromise nobody argues with

The single fastest-moving structure of the year was the credit. Companies in the index offering credit models grew from 35 at the end of 2024 to 79 at the end of 2025, a rise of 126%. Figma, HubSpot and Salesforce all added credits without removing seats.

The appeal is symmetric, which is rare in pricing. The vendor gets a variable cost pass-through that tracks inference spend. The buyer gets a ceiling, expressed in a unit they can put in a budget line. Both sides can sign.

What a credit hides

A credit is a unit the vendor defines and can redefine. One vendor's credit buys a document summary, another's buys a single model call, and neither publishes the conversion to underlying compute. You cannot compare two credit prices, which means competitive pressure on that price is weak.

That matters more as the underlying cost curve moves. Token prices keep falling while total bills keep rising, a pattern covered in the analysis of why cheaper tokens produce larger invoices. Credits sit exactly on that seam, and they are the mechanism by which falling unit costs do not reach the buyer.

The margin pressure behind all of this is real. Poyar's survey found a median target gross margin for AI products of about 50%, against 70% to 80% for traditional SaaS, with only 12% aiming for SaaS-like margins on AI. That gap is the whole reason credits exist, and it is examined in the breakdown of inference costs and what they do to software margins. Salesforce ran the most elaborate version of this play. It added AI credits, new licensing tiers and agentic enterprise agreements in one year, covered in the assessment of whether Salesforce is the platform or the roadkill of the agent era.

Lesson 4: the rollout breaks the change, not the number

The most instructive pricing failure of the period was not a bad price. It was a good price described badly.

In June 2025 Cursor replaced a fixed allotment of 500 fast requests with a usage model offering $20 of frontier model usage per month, plus unlimited use of Tab and of models in Auto. The economics were sound. Users were reaching for expensive frontier models, and a flat per-request price could not survive that.

The failure was a single word

The word was "unlimited". It applied to Auto and to Tab, not to every model, and that distinction was not made clearly. Founder Michael Truell wrote in the company's 4 July 2025 post: "We recognize that we didn't handle this pricing rollout well, and we're sorry. Our communication was not clear enough and came as a surprise to many of you."

Cursor refunded unexpected charges incurred between 16 June and 4 July. Three weeks of revenue, plus a permanent entry in every future competitor's sales deck, all sourced from an adjective.

Most pricing post-mortems blame the model. Here the model was defensible and the sentence was wrong. If you are shipping a consumption change, the test is not whether your finance team agrees the price is fair. It is whether a customer can predict next month's bill from your pricing page without contacting you. Coding tools are where this gets litigated first, which is the terrain covered in the build versus buy analysis for coding agents.

Lesson 5: outcome pricing arrived as a discount

Outcome-based pricing was the most discussed idea of the year and the least adopted. The clearest real example came from HubSpot on 14 April 2026, and it repays close reading.

HubSpot moved Breeze Customer Agent from $1.00 per conversation to $0.50 per resolved conversation, and Prospecting Agent from a recurring monthly charge per enrolled contact to $1 per lead recommended for outreach. Included credit allowances did not change. Chief Customer Officer Jon Dick framed it plainly in the announcement: "You pay when it works, full stop."

Run the arithmetic on HubSpot's own numbers

HubSpot states that Customer Agent resolves 65% of conversations. Take 100 conversations. Under the old price that cost $100. Under the new price, 65 resolutions at $0.50 each cost $32.50.

At the vendor's own published resolution rate, the flagship outcome-pricing move of 2026 is a price cut of about two thirds. That is not a criticism of HubSpot, which is buying adoption with it and says so. It is a warning about how the category is being read.

My position is that outcome pricing is being introduced at a discount because vendors need volume and trust, not because anyone has solved the hard part. The hard part is attribution, and it does not get easier at scale. Expect the price per outcome to climb once measurement is normalised and the buyer has no baseline left to compare against.

There is a growth signal on the other side of the argument. High Alpha's 2025 benchmark data put year-on-year growth at 65% for outcome-priced companies against 34% for subscription-priced ones. The publisher does not disclose the sample size, and AI-native companies are the ones most likely to try outcome pricing at all, so read that spread as association rather than cause.

Company actions are as reported by PricingSaaS and Growth Unhinged, January 2026, and by the vendors themselves for Cursor and HubSpot. The lesson column is this publication's reading, not the vendors' stated intent.

What the year looks like from the buying side

Every one of these lessons has a mirror image for the person signing the invoice. A vendor iterating four times a year is a vendor whose price you cannot forecast from last year's contract.

The scale of the increase is harder to pin down than it should be. Vertice is a spend management vendor with a commercial interest in the answer. It reported SaaS list pricing up 11.4% year on year in January 2025 against 2.7% average G7 inflation, and 58% of vendors raising list prices in 2024. Treat both as directional. The publisher does not disclose sample construction, and a company that sells savings has an incentive to size the problem generously.

What is not in dispute is that the other side of the table is improvising too. SBI Growth found 48% of SaaS pricing leaders still make critical pricing decisions on intuition, rising to 59% among those selling to small business. Set that beside 1,800 changes and the picture is not a market optimising. It is a market guessing quickly.

Three moves are worth making at your next renewal. Ask the vendor for its pricing change history over 24 months, because a vendor that has repackaged twice will repackage again during your term. Ask what a credit converts to in underlying usage, in writing. And require notice terms on unit redefinition, not just on price, because the redefinition is where the increase now happens.

Then measure. The one discipline that survives every pricing model is recording what a tool changed against a baseline you captured before you bought it. That is the argument in the piece on where measurable AI return has actually shown up. Buyers are also facing consolidation pressure from the vendor side, covered in the analysis of point SaaS absorption into agent ecosystems.

Where this argument is weakest

This post rests heavily on one commercial dataset. That is worth stating before you act on any of it.

No dataset tracks whether the changes worked

We know the top 500 companies changed pricing 1,800 times. Nobody has published revenue impact per change. Every claim about which moves were smart, including mine, is inference from what companies kept doing rather than from measured outcomes.

The index also has a composition bias. It tracks companies with transparent published pricing, which skews toward product-led vendors and away from enterprise sales-led ones. The behaviour of the largest software contracts in the world is mostly invisible to it.

The case that this is confusion, not learning

The strongest counter-argument to this entire post is simple. A market that edits its prices 3.6 times a year may not be learning. It may be flailing, because the cost base underneath AI features moves faster than any pricing committee can meet.

I find that case partly persuasive. The honest split is that packaging churn looks like genuine iteration, while the credit wave looks more like a holding pattern that lasts until inference costs settle. If credit adoption stalls in 2026 rather than compounding, the holding-pattern reading was right and I was too generous.

Frequently asked questions

Where to start this week

Run one exercise, on your three largest software vendors, in that order of spend.

Open each vendor's pricing page and compare it against the version in your signed contract. Where they differ, you have found a repackaging you were not told about, and you now have a renewal question with a date on it.

Then do the same to your own price. Write down the last time you moved packaging, as distinct from the last time you moved price. If the packaging date is older, you have been skipping the cheap experiment that 43% of the market ran last year.

Related reading