4 ms·
This is pretty sweet. The speed is nice but what I really care about is you bringing the per token cost down compared with models on the level of mistral mediu
by CuriouslyC 3y ago
This is pretty sweet. The speed is nice but what I really care about is you bringing the per token cost down compared with models on the level of mistral medium/gpt4. GPT3.5 is pretty close in terms of cost/token but the quality isn't there and GPT4 is overpriced. Having GPT4 quality at sub-gpt3.5 prices will enable a lot of things though.
- ukuina 3y agoI wonder if Gemini Pro 1.5 will act as a forcing function to lower GPT4 pricing.
- ComputerGuru 3y agoIs that available via an API now?
- sp332 3y agoKind of, it's in a "Private Preview" with a waitlist.
- sturza 3y agoAnd in non EU countries.
- ComputerGuru 3y agoVia GCP only?
- MuffinFlavored 3y agoWhat's the difference in your own words/opinion in quality between GPT-3.5 and GPT-4? For what usecases?
- CuriouslyC 3y agoGPT3.5 is great at spitting out marketing babble, summarizing documents and performing superficial analysis but it doesn't take style prompts as well as gpt-4 and its reasoning is significantly worse when you want it to chain of thought follow a complex process while referencing context guidance.
- emporas 3y agoMixtral's quality is definitely up there with Gpt3.5. Specifically for coding, i consider them almost equivalent in quality. In fact Mixtral 8x7 is starting to be my go-to coding assistant instead of Gpt. It is fast, it is accurate, and i think i like his responses better than Gpt. Reducing LLM size almost 10 times in the span of a little more than a year, that's great stuff. Next step i think is 3 billion parameters MoE with 20 experts.
- Zpalmtree 3y agoGPT-4 is overpriced vs what?
- int_19h 3y agoYou seem to be implying that Mistral Medium is on the same level as GPT-4?