Today, we’re thrilled to announce an entirely reimagined platform pricing structure designed to provide more flexibility, control, and value to every developer and business building on the Cohere Platform. Our new pricing opens up the platform to allow you greater access and a simpler experience when choosing the right resources for your project. You’ll be able to experiment more for free, estimate costs more easily, and see less jargon from us. Read on for the full details!
1. We’re providing a brand new free developer tier. Natural language processing (NLP) has become part of the public consciousness due to its rapid evolution and increasing number of applications. We want to make it easier for every developer to explore the vast potential of NLP and experiment with it on the Cohere Platform. So, we’re replacing our current $75 free credit program with a new freemium tier that gives you more hands-on experience with our API.
To enable the free tier, we’re introducing two types of API keys that give users access to all of Cohere’s endpoints: Trial API keys and Production API keys. All current API keys will automatically be converted to a Trial API key, which is rate-limited to 100 API calls per minute and cannot be used in production scenarios. If you’re new to Cohere, get your Trial API key by signing up for an account.
For developers looking for higher rate limits, or to serve Cohere to users in an application, you can upgrade to a Production API key where the throughput is 100 times higher (10,000 calls per minute). To upgrade your key now, simply complete the “Go to Production” workflow under the Usage tab in your dashboard (it takes less than 3 minutes).
We hope that the new free tier will allow you to prototype using Cohere without worrying about having to upgrade in order to complete your experiment or proof of concept.
2. We heard you, and we’ve simplified our pricing plans. Thanks to your feedback and our relentless focus on improving efficiencies, we’ve simplified our pricing structure to make it easier for you to estimate the costs of using our platform. We’re also reducing the price of our embeddings (decreased by more than 10x depending on the length of your inputs) to enable and encourage a broader range of applications and tasks to be built with Cohere.
Here’s a summary of the new pricing model per endpoint:
We also want to make it easier for you to use our largest, best performing models by removing the pricing differences by model size — pricing of each endpoint will be standardized across sizes. This way, it’s a no brainer to get the best product Cohere has to offer!
Custom Generate and Embed models will continue to be twice the price of their standard counterparts, whereas pricing for both Classify categories remain the same. Here’s a summary of the new costs:
3. We continue to optimize how developers interact with Cohere. For starters, we’re removing jargon in our product terminology. We recognize that language choices are not only critical to how we communicate in our every day lives, but they also impact the platform experience of our users. One aspect of this is how we refer to our products and features.
Since Cohere began, we’ve been referencing “finetuning” to mean re-training the last few layers of a large language model, so that it performs better at a specific task. To better capture the meaning and spirit of this experience, “finetuned models” will be called “custom models.” This means that instead of referring to the process of “finetuning a model,” you’ll see us referring to it as “training a custom model.”
What about the standard, non-customized models (those that you would use as a starting point for re-training)? They’ll be getting a naming makeover, too. Now, “baseline models” will be called “default models.”
Nothing will change in how you access, use, or train our models, it’s just the terminology that will be different. It may seem like a small thing, but we believe that it will give you a much more straightforward (i.e., enjoyable) experience on the platform. You’ll also notice a number of other changes, including four QuickStart guides for faster onboarding, easier sharing of custom models with team members, accuracy metrics and training logs for custom models, and more! Check out our full release notes.
At Cohere, we’re dedicated to making developers’ lives easier, and we hope these changes enable you to build even more awesome applications with Cohere. Let us know your feedback in our Discord Community or by sending us an email at firstname.lastname@example.org.