>>67043355There's no true 'alternative' in terms of a similar quality level (4 is the best LLM in the world by a fair margin, after all) but you could try:
- 4 turbo (around 90% as good as base 4 and 1/3 the price, but has a few weird prose tendencies)
- a claude proxy (hard to get ahold of but almost as good - better depending on your tastes)
- gpt3.5 (~10x cheaper than gpt4 and decent)
- Mistral-medium (new experimental model, available for fairly cheap through the mistral API - from my testing it's not too far from 3.5's quality and there's zero filtering which is nice)
- Whatever that 70b candidate model is on Agnai's paid tier? (Haven't personally tried this one)
- Mistral-small (very cheap and unfiltered via the same API but definitely a step down - it's around mythomax tier perhaps)
- Mythomax/Mixtral/Llama13 finetunes/etc - copium tier, these really won't feel even close to the good corporate level models.
