For every Bind AI subscription tier, you are allotted a fixed number of compute points which get refreshed every month. For each query you make, compute points are consumed based on the model you select, length of your query (number of tokens in the query), and the length of the response to your query (number of tokens in the response), length of the history or context used (e.g. the history of your previous messages or embeddings which get retrieved.
Bind AI supports 15+ AI models from multiple providers such as OpenAI, Anthropic, Mistral, Meta Llama, Cohere and many others. Each model has different compute requirements, some are computationally intensive large parameter models, while others are more basic. Computationally intensive models are generally more expensive, as higher server costs are associated with it.
Compute points allow you the flexibility to choose the model of your choice based on how many compute points you have available. As an example, if you have less compute points available (e.g. in Lite tier, 200K compute points are available vs Premium tier has 1M compute points available), you can switch to a model such as GPT-4o mini which consumes less points as compared to Claude 3.5 Sonnet or OpenAI O1. Also, it is important to note that each model has its pros and cons, so instead of only choosing a single model for all tasks, it is more effective to pick the right model for the task.
Tokens is a standard measure of unit for LLMs (AI models). One token roughly represents 4 characters. Each query you enter consists of several words (and characters), and it can be measured as tokens. For each token, a set amount of compute points are consumed, depending on the model and amount of data in the input query and output response.
Here is an example sentence: “Hello how are you”.
The sentence in quotes contains 17 characters including spaces, assuming each token is 4 characters, the sentence consists of ~4 tokens.
If you use Claude 3.5 Sonnet for this query, it will consume ~8 compute points (4 tokens * 2 points per token), and if you use GPT-4o mini it will consume 4.8 points. Note that the response from the model and all the history/context included in the query also counts towards the compute points. Bind AI currently does not show compute points used in real-time, and you will get a notification when you consume your points.
See below the compute points usage per token for advanced and basic models. Compute points per token are subject to change based on model availability. Typically, newer versions of models are more efficient and may require less compute points than its predecessors. This is subjective and depends on the size of the model and reasoning capabilities.
You can add your own API keys to get unlimited queries. You can find this option in the navigation where it says “Add API keys”. Currently, API keys are supported only for Anthropic (Claude) and OpenAI models. You can also maximize your usage by picking the models which require less compute points and still have superior reasoning capabilities (e.g. Mistral Codestral is a very competent model for coding, and requires almost half the compute points as compared to Claude 3.5 Sonnet)
Please refer to the Pricing page for more details on the models and limits available for each tier. For any question, please reach out to us via Chat in the bottom right corner (visible in desktop screens), or email us at sam@getbind.co