Pro Tier

tl;dr Today, we are launching our Professional (“Pro”) Tier, a paid plan for individual developers. This will have everything in our free Individual plan plus a series of new features that unfortunately would be too expensive for us to serve completely for free, such as a blazing fast Autocomplete speed, unlimited use of larger models such as Claude 3.5 Sonnet and GPT-4o, uncapped advanced reasoning, and more.

Paid tier for individuals? What happened to the free forever stance? Is this the end to the good times of state-of-the-art AI assistance for free?

We understand the questions, but assure you there really is no doom-and-gloom here. Everything we do is with the sole purpose of driving value to our users and democratizing AI for developers.

Our Free Forever Stance

Almost two years later and the most common question we receive is still “how is this free?” It really came down to our differentiated infrastructure background, which has allowed us to run our system orders of magnitude cheaper than our competitors. We saw an opportunity to translate these cheap costs into a vector for democratizing AI for software developers, and vowed to provide an industry-leading AI tool free for all developers, forever. A broad statement that is still true today. We have no plans on walking back on this promise.

That being said, even in the original blog post where we explained how we could make this decision and why we did, we recognized that there may be a time where even our super-optimized infrastructure would be cost prohibitive for the capabilities that would emerge as LLMs improved. To quote:

In short, we are committed to having an industry leading free tier forever. This includes things like unlimited autocomplete (like today) and anything else we can serve cheaply. We will not nerf this tier just to get individuals to pay. That being said, there may be new capabilities that actually are useful at the individual level that are expensive for us to serve with no limits, and we will wrap that into a Pro tier. But if you don’t want to pay, you have nothing to worry about - we will continue to keep improving your experience as well.

Even back then, we had the foresight that some day, we would need to create a low cost Pro tier to cover the costs for these new features, ones that would be financially irresponsible for us to provide for free. Well, that day is today.

Our primary monetization strategy is still enterprise customers, none of that has changed, but now we will have a paid Pro tier that is still cheaper than anything else on the market for an individual (we still have that infrastructure advantage overall!). This is a tier that even further pushes the boundaries of quality and capability. We have a duty to our users to keep creating the next generation of experiences, and we cannot let costs slow us down.

Pro Tier at a High Level

Let’s get right into what this tier gets you:

The pricing of $10/month, or $8.33/month if purchased annually, is half the cost of any other product on the market, and comes with a higher combined quality and functionality. Since Codeium is available as an extension in every IDE that you already use, it is a no-brainer from an availability perspective as well.

Besides the unlimited use of Autocomplete, Chat, and Command that our free users also get, Pro users also get a series of more compute intensive capabilities, which we will get into details of in the next section:

• Autocomplete Speed: A new blazing fast Autocomplete experience.

• Unlimited Large Chat Models: No caps on Claude 3.5 Sonnet, GPT 4o, LLama 3.1 405B, and more for Chat.

• Uncapped Reasoning: No artificial limits on amount of code that can be indexed, massively expanded use of Riptide as a reasoning backend, and significantly larger context lengths.

As tokens of gratitude for being a Pro user, we will also be adding a Pro-tier badge on your public Codeium profile, as well as giving you the option to join a special role on our Discord.

The New Features

Autocomplete Speeds

Generally, we believe that faster is better. Very low latency has always been part of our differentiation. Especially for Autocomplete, where even just a few fractions of a second could be the difference between a developer seeing a valuable suggestion and not seeing a suggestion at all, if they have typed the next character.

Today, we are announcing a breakthrough in latency, which requires significantly more compute. It is clear when you see Autocomplete speed before and after side-by-side:

Current default speed:

New speed:

This is aptly named Fast speed.

At the same time, we understand that not everyone might want such fast Autocomplete suggestions. Sure, for most people this will just add value in the long run even if it takes a little bit of time to get used to, but in the end, it comes down to user preference. That is why we are keeping the existing Autocomplete behavior as the Default speed, and have even introduced a slower-than-existing Autocomplete experience for folks who have already been worried about Codeium being “too fast,” called Slow speed.

Default and slow speeds are both available on the free Individual plan, with Fast speed available on the Pro plan, and can be selected in your User Settings panel:

Unlimited Larger Models

We already announced Claude 3.5 Sonnet, GPT-4o, and Llama 3.1 405B integrations into our Chat functionality last month. We gave these models to all of our paying Teams and Enterprise SaaS users because, again, giving unlimited access to large models for free would be financially irresponsible, especially with APIs that we cannot optimize under the hood. So, very simple, let’s just give it to all Pro users too:

Uncapped Reasoning

This one is less visual, but arguably one of the most important. One of the main reasons why Codeium has been adopted by thousands of enterprises, including many Fortune 500 companies, is because of its ability to reason about existing codebases to give responses that are personalized and grounded in what already exists. No more generic answers from a good, but generic, system.

We have built many systems, such as our context awareness engine, invested in research into reasoning breakthroughs, such as Riptide, and trained models with larger and larger context lengths to be able to execute on this reasoning. However, these do not come for free, and are significantly more expensive from a compute perspective than much more naive approaches that users get from other tools.

The free Individual tier gets access to all of these, but only as compute availability allows for. Today, we dynamically scale up and down the execution of the advanced and expanded versions of these capabilities based on compute. With the Pro tier, you will always get the best reasoning capabilities: no limits on indexing, context awareness, Riptide, context lengths, and more. The best quality responses, all the time.

It is also important to mention that to continue to support these additional capabilities for Teams, we have increased the pricing for Teams from $18/user/mo (or $180/user/year on an annual plan) to $24/user/mo (or $228/user/year on an annual plan). For existing Teams customers, we will not increase the pricing on the next billing period, so you will be grandfathered in with existing pricing, as a token of appreciation for being early customers.

Looking Forward

So, what is going to happen with the free forever plan?

Well, we hope it is clear that the free Individual plan isn’t going anywhere:

• If we can serve a new capability cheaply, then we will give it to everyone on the free Individual plan, just like we have given unlimited usage of all of our core modalities.

• If we can optimize some of the features that are today unique to the Pro plan so that they become cheap enough for us to give to everyone on the free Individual plan, then we will just move it to the Individual plan.

• If we can give a limited number of uses of these more expensive features to everyone on the Individual plan, we’ll find a way to do that.

• If we have extra unused compute capacity at any moment, we will use it to serve things such as Riptide, expanded context lengths, and advanced reasoning to our Individual users, without them even knowing, because we always are trying to give the best experience within basic financial constraints.

All of this to say, it doesn’t matter what version of Codeium you use as an individual developer - it will be more powerful than anything else on the market.

In fact, our Pro tier will actually allow us to move even faster. Now we can release features to a subset of our users without needing to spend even more R&D to optimize them to a point that we can run it at 700K+ developer scale in a financially responsible way (and that is assuming that it is possible to do so).

As a preview on how fast this allows us to move, take a sneak peek at Supercomplete, the next capability that will be coming to Pro. Autocomplete is great at predicting the next characters. Supercomplete predicts your next intent, which could be to refactor or to delete existing code, not just adding more code wherever your cursor ended up. Coming to VSCode very soon:

We would love it if you were to try the Pro tier, since we believe it is yet another step in the evolution of AI for software developers.