Ask HN: AI Cloud Computing, is it cheaper than the OpenAI API in the end?

[calipsow]

The question assumes to use the Llama 3.1 405b model.

When it’s cheaper via cloud (per token), which provider would you recommend to me, with focus on saving much money possible?

[atilla_bilgic]

If you will be training a model for a specific purpose you should definitely be using an architecture fit for the purpose. But if you would like to experiment with the most recently released models you can go to https://chatllm.abacus.ai/JBDrthHYkz (Disclaimer: it is a referral link) and create an account for 10$/month without worrying about the number of tokens you will use. Or for the pay-per-use model, you can check out https://openrouter.ai/models for the really rich catalog of models available.

[calipsow]

Thanks for the resources I’ll check them out, my purpose behind this is actually I want to build an app, which requires to call a LLM to serve the features I want to add, so I thought either build my own http api wrapper around an LLM, or just use the OpenAI api.. it depends on what’s cheapest in the end. And I’m not planning to train any models, just running existing open source models through a cloud provider.

[JSDevOps]

The two are completely different things. Depends what your requirements are.

[calipsow]

Im looking for an solution for one of my apps that need an LLM to serve the features I want, so either I write an http wrapper for an existing LLM and run it through the cloud, or I just use the OpenAI API for my app.