AI SaaS Pricing Weekly: 12 Price Changes, Cursor Up 50%

AI SaaS Pricing Weekly: 12 Price Adjustments, Cursor Up 50%

Pricing for AI and SaaS products is entering a period of high-frequency adjustment. A monitoring project that continuously tracks 133 pricing pages recorded 12 substantive changes this week: 2 clear price increases, 1 clear price decrease, and the rest structural reorganizations or free-tier contractions. Although this sample is small, it reflects several real signals about the commercialization of AI products today—inference costs are falling, but end-product prices have not moved down in tandem; free tiers are being systematically tightened; and the pricing unit itself is shifting from "seats" to "outcomes."

The Two Increase Samples: Cursor and Activepieces

The only two products explicitly marked as "price increase" this week come from AI coding and automation, respectively, and both increases are significant.

Cursor: Entry Price Jumps from $40 to $60

Monitoring data shows that Cursor removed the $40 / user / mo tier while adding $60 / mo and $200 / mo. Calculated on the lowest available tier, the entry price rose 50%. The appearance of the $200 tier also indicates that the product line is extending upward, targeting high-usage scenarios for power users or teams.

Coding assistants have been one of the most fiercely competitive categories over the past year. By raising prices now, Cursor is essentially testing the upper limit of users' willingness to pay for "AI productivity"—when autocomplete and Agent capabilities become essential, price elasticity will be noticeably lower than for ordinary SaaS tools.

Activepieces: $16 to $20, a 25% Increase

Open-source automation platform Activepieces adjusted its price from $16/mo to $20/mo, with the corresponding annual tier changing from $166 to $200. The newly added selling point description is "3× the Free plan, no daily cap"—removing the daily call limit is a substantive capacity expansion for automation workflow users. This round of price increases looks more like a hedge against pressure on the free tier: using a higher paid tier to accommodate users who max out their quotas.

Free-Tier Contraction: Supabase, Deepgram, and Murf AI

Changes to three products this week point in the same direction—free quotas are shrinking.

Supabase Directly Removed the Free Plan

The monitoring record is very concise: removed: Free Plan. For Supabase, which has long used a generous free tier as the core of developer acquisition, this is a landmark move. The free tier of a developer platform is never charity; it is customer acquisition cost. When the user base is large enough and the conversion funnel mature enough, tightening the free tier is an inevitable business choice.

Deepgram: Limited-Time Offers End, Prices Return to Standard Tiers

Deepgram removed multiple transitional prices in the form "through 9/14 Then...", such as $0.041/min through 9/14 Then $0.059/min. The cleanup of this "return to original price after the promotional period" structure means new users no longer enjoy early discounts. The newly added multiple per-minute and per-character billing tiers (such as $0.059/min, $0.075/min, $0.0405/1k characters) show that its billing dimensions are becoming more granular.

Murf AI: Subscription Tiers Replaced by Usage-Based Billing

Murf AI removed subscription tiers such as $0/month, $19/month, and $66/month, and added per-character billing options such as $0.01/1000 characters. The disappearance of the free tier and the switch in billing unit occurring at the same time is a typical transitional form of "moving from subscription to usage-based."

The free tier is a common customer acquisition method for SaaS and developer tools: attracting developers to try the product with a limited free quota, then guiding paid conversion through usage caps or feature restrictions. In AI products, the cost pressure of the free tier is especially prominent—every inference call incurs real compute expenditure, rather than the near-zero marginal replication cost in traditional SaaS. Therefore, as the user base expands, the free tier often shifts from a "customer acquisition investment" to a "continuous bleeding point." This week's removal of the Free Plan by Supabase and deletion of the free tier by Murf AI are precisely manifestations of this logic: free-tier contraction usually occurs after a product has completed market education and its conversion funnel has matured, and is a signal that commercialization is moving from the expansion phase into the harvesting phase.

Structural Reorganization: AssemblyAI, Fireworks AI, and xAI

Several companies did not simply raise or lower prices, but restructured their pricing methods. Such changes often have a greater impact on developers' actual costs than headline prices.

AssemblyAI added tiers priced per million units, such as $0.20 / 1M and $12.00 / 1M, and adjusted the tier descriptions in the low-price range. Fireworks AI removed the two large-amount tiers $18.00 and $7.00, replacing them with small-amount pricing in the range of $0.134 to $0.334—moving from "monthly/volume packages" to more granular per-call billing.

xAI's change is the most直观: Speech to Speech changed from "Starting at $0.05 / min" to $0.08 / min, a 60% increase, and the vague "starting at" wording was removed.

Only One Price Decrease: Anthropic

The only product marked as "price decrease" this week is Anthropic. The changes are concentrated in the description of Claude's Pro plan: US$20/seat/month, billed annually was added, while scattered price points such as $1.50, $18.75, and $75 were removed.

This is closer to a cleanup and unification of the pricing page than a broad price cut. But viewed together with the price trends at the infrastructure layer such as Deepgram and Fireworks, a judgment can be drawn: the marginal cost of model inference is falling, but this benefit currently remains mainly at the API layer and has not yet been fully passed through to the subscription prices of end SaaS products.

The Pricing Unit Is Shifting: From Seats to Outcomes

The most noteworthy detail this week is hidden in the changes at Intercom and Asana.

Intercom removed $29per agent/mo and From $99/mo, and added $29per teammate/mo, $119.09/mo, $90.09/mo, and $0.99per outcome. From "per seat" to "per outcome," the billing anchor has changed from headcount to the amount of work AI actually completes.

Asana's move is similar: it removed $5.99 per user, per month and multiple AI Studio credit packages, replacing them with $0.50 per AI request, billed upfront and $0.60 per AI request, billed monthly. The unit price difference between prepaid and monthly billing ($0.50 vs $0.60) is a typical usage-based pricing technique.

When AI begins to replace some manual operations, charging by seat loses its rationality—a team may have few people but a large volume of AI calls. Billing by outcome or by request is becoming the default pricing language for AI-native features.

Per-seat pricing is the most mainstream pricing method in traditional SaaS: charging by the number of user accounts using the product, with price linearly tied to team size. Its rationality lies in the fact that the value of software increases with the number of users. But AI-native features break this assumption—a three-person team may generate massive AI calls, while a hundred-person team may barely use AI features. Per-seat pricing fails here: it can neither reflect real compute consumption nor capture the value created by AI. Outcome-based pricing places the billing anchor on the work AI actually completes, such as Intercom's $0.99/outcome and Asana's $0.50/AI request. This model directly aligns revenue with the value delivered by AI, but it also brings challenges such as reduced revenue predictability and ambiguous definitions of outcomes. It is currently a frontier experimental direction in AI commercialization pricing.

What This Week's Data Shows

Putting these 12 changes together, three trends can be distilled:

First, free tiers are becoming a scarce resource. Supabase, Deepgram, and Murf AI all tightened free quotas or unlimited tiers at the same time, indicating that the marginal cost pressure of AI products is real, and the early strategy of "burning money for growth" is retreating.

Second, price increases are concentrated at the application layer, while price decreases appear at the model layer. Cursor, Activepieces, and Pika raised prices, while Anthropic lowered prices. The application layer is fighting for profit margins, while the model layer is fighting for developer share.

Third, pricing granularity is becoming finer. Per minute, per character, per million tokens, per request, per outcome—all products are looking for measurement units that more accurately reflect cost and value. Coarse-grained monthly subscriptions are being gradually dismantled.

It should be noted that this monitoring only captures changes in price-related text lines; page layout and copy changes are not counted, so the above 12 are all real pricing content changes. The sample size is 133 pages, with 12 weekly changes—about 9% of pages underwent price-related adjustments within seven days. This frequency itself shows that AI SaaS pricing is far from stable.

The price divergence between the model layer and the application layer stems from the difference in their cost structures. The model layer (such as Anthropic's API) directly faces the marginal cost of inference compute. As model efficiency improves, chip supply improves, and competition intensifies, unit token costs continue to decline, and price cuts are a direct means of competing for developer share. The application layer (such as Cursor and Intercom), on the other hand, adds product R&D, engineering integration, and user experience investment on top of model costs, and its pricing depends more on users' willingness to pay for "AI productivity" than on underlying costs. When the application layer faces fierce competition, raising prices is instead a strategy to test the upper limit of value and screen for high-value users. This scissors gap of "upstream price cuts, downstream price increases" means that the model cost dividend is currently being retained by the application layer as profit margin and has not yet been passed through to end consumers.

One-line Summary

12 changes across 133 pricing pages this week reveal three major trends: free-tier tightening, application-layer price increases, and a shift to outcome-based billing.

Paragraph Summary

A monitoring project tracking 133 AI and SaaS pricing pages recorded 12 substantive changes this week: Cursor's entry price rose 50%, Activepieces raised prices 25%, and only Anthropic showed a price decrease; Supabase removed its free plan, Deepgram cleaned up promotional prices, and Murf AI deleted its free tier, with all three tightening free tiers simultaneously; Intercom and Asana changed billing units from per seat to per outcome or per AI request. Overall, three trends emerge: free tiers are becoming a scarce resource, price increases are concentrated at the application layer while price decreases appear at the model layer, and pricing granularity continues to become finer. About 9% of pages underwent price adjustments within seven days, indicating that AI SaaS pricing is far from stable.