Welcome back to Good Better Best.
Each week, we break down real pricing, packaging, and product moves from SaaS and AI leaders and share the ideas worth stealing.
This week, we’re digging into one of the most common issues in SaaS today: how to bridge the gap between subscription and usage models. OpenAI and Anthropic just made matching moves to drive subscribers to use their API. Below, we broke down why, and what SaaS and AI leaders can learn.
Before we get there, we have two great events coming up. Grab a seat below:
▶ Thursday, November 12, 12:30 PM ET: Office Hours with Ulrik on migrating to credits. Register here.
▶ Thursday, November 19, 12 PM ET: AMA with Jas Garcha (Head of Pricing at Vercel) on how Vercel is monetizing agents. Register here.
Let’s get to it.
This Week in Pricing, Packaging, and Product
This week we observed 324+ changes. The highlights:
OpenAI added monthly API credits to ChatGPT Pro, matching each tier at $100, $200 or $500, with Free, Go and Plus getting none [Link]
ClickUp launched a new top-tier MAX plan at $100 a month and folded Brain AI and Everything AI into pooled AI credit allowances on every core plan [Link]
Stripe repriced Global Payouts to 50 free domestic payouts a month then per-rail fees, and added a $15 dispute appeal fee [Link]
Datadog replaced its Free tier with a Basic tier that uses contact-sales pricing, 15-month retention, and no container monitoring [Link]
IFTTT moved to credit-based tiers [Link]
Veed raised Creator to $12 a month billed yearly ($144 a year) and Pro to $22 a month ($264 a year), added a Studio plan at $39 a month billed yearly, and raised Pro AI credits from 18,000 to 30,000 a month [Link]
Mem0 replaced its four-tier lineup with Free and Custom, removing self-serve Starter at $19 and Pro at $249 [Link]
Sprout Social added Trellis Plus at $28 per seat per month for 1,000 AI credits, while base Trellis keeps 100 free credits [Link]
Ahrefs added Traffic Analytics as a new product, with Traffic reports at $99 a month and Traffic and Market reports at $299 a month [Link]
GitLab launched Orbit as an add-on for Premium and Ultimate, sold in query packs paid with reserved GitLab Credits [Link]
GitHub added Code Quality, Code Security and Secret Protection as standalone add-ons [Link]
Vercel added usage-priced Vercel CI at $0.0035 per CPU minute after 200 included minutes, and Custom Monitors at $0.0035 per GB scanned [Link]
Synthesia removed the Roleplay Training learner-seat add-on at $25 per learner per month yearly [Link]
Canva changed Free AI usage from up to 20 Standard or Premium AI uses to limited access, and dropped the usage estimates from the footnote [Link]
Taskade added a Starter plan at $6 a month billed yearly for 3 users, between Free and Pro [Link]
Speechify started showing Premium at $139 billed yearly alongside $29 a month [Link]
Zoom added a Workplace add-on at $49 a month for SIP and H.323 conferencing endpoint support [Link]
Firecrawl changed Map billing from 1 credit per page to a flat 1 credit per Map call [Link]
Krea removed LoRA training from Free, while paid plans and prices stayed the same [Link]
Replit dropped Plan mode from Core, and dropped 10 parallel agents and Invite up to 50 viewers from Pro, with prices unchanged [Link]
Runway added App Studio to Standard, Pro and Max, and launched Kling 4.0 with 25% off through Nov. 30 [Link]
Gusto added General liability, Professional liability, Business Owner's Policy and Commercial property insurance add-ons priced on premiums only [Link]
Check out more updates on PricingSaaS →
The problem: How do you get subscribers on usage pricing?
If you sell AI inside a flat monthly plan, you have a problem.
Everyone pays the same price, but nobody uses the same amount.
Light users are very profitable.
Heavy users can cost more to serve than they pay.
Raise the price for everyone and you lose the light users who were paying for the heavy ones.
The labs know this better than anyone. Last month, SemiAnalysis published rough estimates of Anthropic's numbers:
By its estimates, here’s how Claude subscriptions and usage break down:
Share of Compute: Subscriptions (42%) vs API (58%)
Share of Revenue: Subscriptions (10%) vs API (90%)
Gross Margin: Subscriptions (50%) vs API (92%)
These numbers (even if directional) make it very obvious that it pays to shift more users to usage based pricing. The question is how?
Most companies try one of two moves.
Cut the limits. Give each plan less usage for the same price (aka SaaS Shrinkflation).
Force a switch. Replace the flat plan with pay-per-use.
OpenAI cut the value of its $200 ChatGPT plan in half this fall, before reverting after pushback. SemiAnalysis also says backlash has forced Anthropic to walk back planned limit cuts more than once.
Forcing a switch feels straight up hostile, especially right now as customers are increasingly looking for predictability.
What the labs did instead: API Credits for select subscribers
Instead of shrinking the subscription, the labs added to it.
On October 7, Anthropic started giving subscribers free monthly API credits. On October 28, OpenAI did the same.
▶ Anthropic: Max ($100 a month) gets $100 in credits. Ultra ($200) gets $200. Team plans get $20 per Standard seat and $100 per Premium seat, pooled and capped at $500 a month.
▶ OpenAI: ChatGPT Pro's $100, $200, and $500 tiers each get the same amount back in credits. Business gets $20 per Standard seat and $100 per Premium seat, pooled and capped at $500 a month.
Credits expire every billing cycle and don't raise chat limits. The cheaper plans (Claude Pro, ChatGPT Go and Plus) get nothing.
Importantly, nobody's plan got worse. Nobody was forced to switch. The best customers just got free money to start using the higher-margin product. When the credits run out, the next dollar is a paid one. And it costs the labs less than it looks, because credits that expire won't get used.
Through this lens, it’s explicitly clear that the subscription is the acquisition strategy, while the API is the expansion strategy.
You've seen this model before
Here's the funny part. The labs borrowed this from SaaS.
The most common SaaS model right now is a seat or license, plus a credit allowance for AI inside it, with more usage for sale on top.
▶ GitHub Copilot Pro+ costs $39 a month and includes $39 in AI credits.
▶ HubSpot includes HubSpot Credits, which pay for its AI agents, with each edition.
▶ Notion's Business and Enterprise seats come with AI credits built in.
Usually, we’ve seen pricing ideas flow from the labs down to SaaS pricing pages. This time, it went the other way.
What you can Steal
If heavy users are eating your margins, here's the short list of action items to consider:
Keep the subscription. Most customers like a predictable bill. Don't rip it up or force everyone onto usage pricing.
Have a clear pay-as-you-go option outside the subscription. That could be a set price per credit, credit packs people can buy in varying volumes, or volume discounts that kick in as users buy more. Whichever you pick, customers should know exactly how to buy more usage and exactly what it costs.
Set included credits thoughtfully, plan by plan. This is the hardest part. Aim for healthy margins while making sure you never lose money on your power users. When Clay launched its new usage meter for actions, it set the limits so 95% of customers wouldn't hit them on their current plan. You don't need to be that generous, but it's a solid guideline.
Make the next dollar easy. When someone hits a limit, buying more should take one click, not a sales call.
As a first step, pull your top 10% of users by AI usage and write down what each one costs you to serve compared to what they pay. If any of them cost more than they pay, that's where your limits or your usage option need work.
If you could use a sounding board or another set of eyes, you know where to find me.
