Fireworks AI is one of the most overpriced model hosting companies in the ecosystem. I tried their fine-tuning service with a very small dataset of 10k records using SFT with gpt-oss-20b.

It ran for 8 mins, and I got a bill for $192 !!!

You can get A100s for $1.29/hr on Lambda, for 8 minutes of time that’s $0.17

This is maybe the biggest markup I’ve ever seen in the industry. Anyone can just spin up a runpod or modal instance and do it for a small fraction of the cost.

Anyway, buyer beware, this is money I won’t get back. Not to mention the service is overall just mediocre.