When we launched the Windsurf Editor with Cascade, we had no idea whether people would find value from the tool or not. We thought Cascade was revolutionary, but we also didn’t want to overcomplicate pricing assuming that usage was going to be insane. Instead, we just decided to give everyone a free trial with a limit of 1000 steps per month to make sure costs don’t go out of control. Steps were more of an internal metric that was roughly the sum of the number of user prompts and flow actions, but really, the important note here is that none of us internally at Codeium got close to the 1000 steps per month while testing pre-launch. We thought we were giving a very generous plan.
Let’s just say that in the last couple of weeks we have learned that we are not representative of all developers. It turns out that a lot of developers actually did get lots of value from Cascade, that it was the step change in AI that we believed it to be, which is super exciting for us. And let’s also say that a lot of developers blew through the 1000 steps in the matter of days. We knew that we hadn’t exposed a “steps number” visibly to users, so we made the internal call to completely scrap the rate limiter under the hood and not cap usage to anyone until we figured out a better long-term strategy. That was just the right thing to do by our users, even if that meant it cost us a lot of money. Like, a lot.
But of course, that is not sustainable, and we are not fans of the strategy to burn VC money without having a real plan to sustained revenue.
Going under the hood, the expensive part of Cascade is the usage of premium models like Claude 3.5 Sonnet and GPT-4o to perform some of the reasoning that is applied to both user prompts and flow actions. They are not the only models that are used within Cascade, but we have built the other models ourselves and can optimize the serving infrastructure a ton. We can’t do that with these premium models.
So, two things had to change — one, we had to put in a credit system for real for these premium models and two, we had to have a backup to these premium models. On the latter point, we have that Cascade base model now. So, let’s talk about the premium model credits.
We are changing the pro tier to be $15/mo, but instead of 1000 steps, users will get access to 2000 steps — 500 User Prompt credits and 1500 Flow Action credits. Every time a user uses a premium model to prompt Cascade, that costs one User Prompt credit, and every time the AI flow performs an action, that costs one Flow Action credit.
As mentioned earlier, once the user runs out of premium model credits, the Cascade Base can be used, or there will be an option to purchase more credits. We’ve learned from hundreds of thousands of developers using Cascade that there is wide variance on whether they would need more User Prompt credits or Flow Action credits, so these additional purchased credits are referred to as Flex credits, and can be used to backfill for either User Prompt or Flow Action credits. Any purchased Flex credits will roll-over to future months indefinitely.
On the Pro tier, users can purchase 300 Flex credits for $10. We know that that might sound steep, but the reality is we have been constantly hitting capacity limits with folks like Anthropic. There’s not enough GPUs. This wasn’t on our bingo card either.
That being said, we do know a large fraction of developers do want to use a lot more. So, we have decided to roll out a brand new Pro Ultimate tier. The Pro Ultimate tier is priced at $60/mo, but with it the user gets an infinite bank of User Prompt credits and 3000 Flow Action credits. And for any additional purchase of credits, the user will get 400 Flex credits for $10 instead of 300 like on the Pro plan. Pro Ultimate is the cost effective plan for any power user.
At this point, we recognize many people will say something along the lines of “let me bring my own API key or pay per token and forget all this credit system nonsense.” We believe there may be some underestimation on how many tokens we are processing on every reasoning step through the combination of knowledge retrieval, encoding of developer actions, and long history, all to give the magic that users feel with Cascade. We are only a couple weeks post-launch and already processing hundreds of billions of tokens with just Claude 3.5 Sonnet every single day. We get nice discounts from these API providers and pass all of the savings from those discounts to our users. So, it is probably not a great idea for users to take these costs on themselves without these discounts. In fact, we have expertise and confidence in our ability to continue to bring costs down over time, and plan on continuing to pass those savings back to users in the form of more credits at every price point.
For those new to the Codeium journey, there might be questions on how we plan on making money if we pass all the savings back to the developer. Unlike other AI IDE products, we have an entire B2B Enterprise arm to our company that is highly profitable. We hit eight figures in annual recurring revenue (ARR) for that business in less than a year. So unlike other AI IDEs who are trying to make money off the individual developer, that is just not our profit-making target. Just because we have seen success with Cascade doesn’t mean we are changing the pricing to now make quick money, we are just trying to make sure this B2C side of the business doesn’t financially sink us.
Of course that might just sound like what a vendor would say, but just look at the fact that we’ve kept our promise of a free tier IDE extension for years. The Codeium extension for other IDEs has been better than paid extensions like GitHub Copilot for a long time. We do lose money there but it’s fine because we’ve been able to keep costs under control.
Wrapping up — yes, we are changing pricing and creating a credit system. And yes, we understand this might not be the most amazing news. We’re not fans of credits or caps either. But also yes, we do want to keep pushing the edges of what is possible with AI and flows without having to worry whether a tweak in how we do things will drive us into financial ruin. We are humbled by the fact that a large number of developers are getting the value from Cascade that we hoped it would drive, and simply just want to get back to work to build the best AI.
Our goal at Codeium has always been to let every developer dream bigger. To build things bigger, better, and faster than they could before. And seeing some of the things that have been built in the last couple of weeks, with stories of how Cascade felt like magic or people without coding backgrounds showing off their first app? All of this just gets us more fired up to make the AI even better.
Surf’s up.
