The French start-up Mistral has released LLMs that some have claimed are close to GPT-4 in performance. The company is providing only downloadable models and an API at present, neither of which is very easy to work with. But I had some time on my hands, and with GPT-4’s help I was able to make a webpage for interacting with it:
I tested it with Japanese-to-English translation and a few other tasks, and the results seemed on par with GPT-4, Bard, and Claude.
Depending on how much you use it, it might be cheaper than the other for-pay options. I have used 75,564 tokens so far, mostly with the more expensive mistral-medium model, and my bill now comes to 0.28 euro.
Tom Gally