I now have access to Mistral's API

Access to Mistral’s API

Just under a week ago, Mistral announced that they would provide an API as a service just like OpenAI and other big names in the world of artificial intelligence. I didn’t hesitate for a minute to sign up for the waiting list and today, I have finally gained access.

Onboarding Process

The onboarding process is very straightforward. Mistral asks for your tax information and uses Stripe as a payment gateway. If you’re already a Stripe user, as I am, the process is even faster. This user-focused approach makes starting up virtually frictionless.

Functionality and Costs

Like other platforms, Mistral charges for usage and allows you to set a spending limit (the default limit is €150).

As for the API, they currently provide 3 main endpoints:

1. Autocomplete text: This is the best-known endpoint, where you can send prompts with the type (system, user, or AI), with the typical OpenAI parameters like temperature, random seed, max tokens, etc.
2. Generate embeddings: This endpoint allows you to generate embedded representations of texts.
3. Query models: This endpoint lets you consult the available AI models.

Available Models

Here is where a little disappointment came in. The available models for now are the “small” ones, that is, the famous Mixtral (Mix of Experts) is not available. However, this is understandable since we are in the early stages of the API launch.

The three models available right now are: tiny, small, and medium. The respective prices are €0.14/1M tokens, €0.6/1M tokens, and €2.5/1M tokens for input, and €0.42/1M tokens, €1.8/1M tokens, and €7.5/1M tokens for output.

Initial Tests

I plan to run some tests with the agents as soon as I have a moment. For now, the “hello world” (example provided by them) has given me the impression that it took a bit longer than expected, but it’s still too early to draw any conclusions.