Free for testing LLM API

Axel ULLERN

Hi, do you know any of the LLM providers that proposes a free API with free prompts through Dataiku LLM mesh (on limited basis of course) so using one of the LLM recipes of Dataiku

the reason why i ask this : i am teaching to students data science using Dataiku , i would like to have my students practicing Gen AI through Dataiku and my students are not ready to pay for accessing LLM and they don't understand (me either ) why ChatGPT or others accept a free direct access (ask a question ) but when you access through a platform like dataiku with an API key it is not free anymore.

is there a technical reason that justify this difference of tarification according to the access ? Why not a free limited number or size of prompts ?

I made some research : the french based LLM from Mistral seems to propose a free API key but when I want to access Mistral (with the API key created on Mistral website )from a LLM recipe of dataiku i am blocked : i suppose because i did not provide my credit card number to Mistral ? then i was expecting that DeepSeek because it is open source LLM would propose a free API plus limited number of prompt , but apparently not.

Thanks for your help

Axel

Operating system used: Mac or Windows

Alexandru

Regarding the costs associated with the course, please note that the tutorial below was specifically optimized to ensure low LLM costs by using a reduced dataset: https://academy.dataiku.com/path/genai-practitioner/text-processing-with-llms-using-visual-recipes/2044161

Various LLM providers offer free tiers with API access. However, these free tiers often come with strict rate limits. To successfully run recipes within these limits, we recommend:

Tuning connection settings, adjusting retries to handle occasional rate-limit errors, and reducing concurrency

Dataiku supports a wide variety of LLMs. You can view the supported connections here: https://doc.dataiku.com/dss/latest/generative-ai/llm-connections.html

Please note that provider pricing and rate limits are subject to change; therefore, we recommend verifying them directly with the provider.

That said, if you use the cheaper models and the reduced dataset mentioned in the Academy tutorial, your token costs should be very low.