6

The quality of Mixtral 8x7B is perceived as better than chatgpt-3.5 and Claude Instant[1], and because it's open-source, there is a lot of competition and you can find providers who say they can have a throughput of 100tokens/s at 0.6$/1M tokens[2]. This is faster, cheaper and better than Claude Instant.

You should also check https://openrouter.ai/, which has very good prices and the possibility of picking closed-source or fine tuned models, but in this case, the throughput for the Mixtral is not as good as at together.ai.

[1] https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard
[2] https://www.together.ai/blog/mixtral

Choosing the model is welcome, but I'm guessing Mixtral will just be strictly better and faster.

    FWIW they already have mistral medium as an option in kagi assistant.

      15 days later
      No one is typing