Mistral AI has just announced Mistral Large, it’s new frontier model.
It’s still behind gpt-4 on every comparable benchmark that I’ve seen, but it’s also $0.008 per 1K input tokens and $0.024 per 1k output tokens, making it 80% the cost of gpt-4-turbo.
If anyone gets the chance to test it out and compare against gpt-4 please share your results!
Is it available? I have a mistral acct and have been using mistral-medium for a couple months.
So far, I’ve not been able to make a successful API call to mistral-large-latest.
.
(
[message] => The upstream server is timing out
[request_id] => 6729855dd5615e06afb5d17cdbf81861
)
Have a test here…
In waiting list. Looking for feedback. Thank you
So far its great, but gpt4 easily beats it.
It is far better at reasoning and complex problems then any open model we have seen before.
In terms of api turbo is still faster and generates more accurate results where mistral can wander off on some cases so the extra 20% is worth it in my opinion.
But competition is always promising 
My use-cases value reasoning and complex problem solving over everything else, so I’ll definitely be sticking with gpt-4-turbo.
It’s a bit of an odd niche, not quite sure who benefits from such a similar price with noticeably worse “intelligence”.
Uh, are there people who don’t have access to the mistral-large-latest API yet?
I only heard about Mistral AI yesterday and was able to use it immediately after creating an account (signing up) and registering my credit card.
I feel sorry, it seems unfair that people are still waiting.
Open source.
While they may not be as good, they are almost entirely open source, and (probably not anymore since Microsoft partnershiped with them) the new large model could be too.
Many people support Godot over Unity,
But Unity the closed engine will always produce more successful and groundbreaking games then the open engine godot.
Medium and Large are not open, only mixtral and mistral. Not sure if this will change or not, but for now the flagship models by Mistral are closed.
I’ve been using Mistral Large to see how my agent prompts would ‘react’ when I inject them into the LLM.
The inferences seem on par with GPT-4 at first sight with I sligthly better ‘understanding’ of the instructions.
No changes needed to be made to existing OpenAI prompts.
I will need to do some more tests, although this Mistral Large models seems very promissing.
have you actually tried mistral large? from my experience it can outperform gpt4 API in some cases
Yup, and it was extremely impressive, but not quite as good at the niche tasks I need. (I’m in an industry with relatively little public data)
Agreed that it can be better for some use-cases, but from what I can tell that’s down to how they fine-tuned the model, not because it actually knows more than GPT-4.
