Have you tried the chat? It gave me so many nonsense answers that I will not be returning any time soon. Putting themselves adjacent to Chat GPT 4, or even putting them on the same page as Chat GPT is absolutely dishonest.

https://chat.mistral.ai/chat

Am I using these wrong? I asked a couple git and python questions and it answered it about the same as GPT-4 Turbo (or whatever ChatGPT uses nowadays). The answer was slightly better than GPT-3.5 Turbo in the sense there was a lot of fluff in the GPT-3.5 Turbo's answer.

I tried it. Seems fine. What prompts gave you nonsense?

I tried what Google would call "long tail" queries. Chat GPT-4 gave me accurate answers, but Mistral gave me nonsensical answers. I can't share the exact prompts because they are personally identifying.

And you can't share them and replace the personal information with xxx?

Which size model are you using? Large isn't terribly good, but Next is alright. It's not close at all to GPT-4, but I can see some use cases I'd try it for (and will be).

I haven't tried this model yet (they seem to be under high load). Mistral's previous top model (Mixtral 8x7B) wasn't quite as good as GPT-4, but it seemed close. Clearly way better than GPT 3.5. Many Mixtral responses were better than the GPT-4 ones.

They compare to Gemini Pro 1.0...

Seems intentionally misleading.

Right. Gemini Pro 1.5 scores 81.9% on MMLU and is also above in a few other benchmarks.

Which - importantly - is better than Mistral at 81.2% ...

Gemini Ultra scored 90% which is better than GPT-4.

This reads like a paid-for press release from Microsoft to pretend like they're almighty and Google is incompetent.

Didn't Gemini 1.0 Ultra turn out to be worse than GPT 4 Turbo in practice? Synthetic benchmarks aren't a perfect representation of anything.

I think they're close enough that it depends on your use case. IME, which is mostly generating Python and shell code, Gemini Advanced is faster, less lazy and generates overall better solutions than GPT-4. It also seems to be more up-to-date with libraries and stuff. It can also directly go to URLs (e.g. summarise a paper), which GPT-4 refuses to do.

Doesn't MMLU have like 2-3% wrong answers anyways ?

With 128K context (1M paid) compared to 32K. Man, Google is going to be a game changer for especially free AI tiers.

Edit: BTW, more Mistral benchmarks here: https://docs.mistral.ai/platform/endpoints/ TIL Mistral Small outperforms Mixtral 8x7B.

1.5 was probably released too late to be tested.

Is 1.5 even publicly available yet?

Yes

How do I use it? I only have access to 1.0 Ultra through Gemini Advanced.

Me too, I don't think it's public yet.

And that doesn't count. Only pro 1.0 available for me through APIs. I need to be able to test for myself the capabilities.

As it stands best LLM available by API by Google is far behind GPT4.

it’s not public, there is waitlist

But Ultra 1.0 is available to compare against, right?

How can I get access to it? The best I can compare against is pro 1.0.

Most of my usecases are logic based on embedded content in the prompt and nothing available to me beats GPT-4 there.

Ultra 1.0 is theoretically available via API, but it's only available to allowlisted customers, which seems to be a very small number right now. It's not generally available.

Ultra 1.0 is the model behind Gemini Advanced, which is generally available.

Not via an API, which is needed to perform proper tests.

From the article:

> generally available through an API (next to GPT-4)

But think of the poor developers...they won't get promoted unless they show impact by being in the lead.

it sounds like they are trying to be clear they aren't stepping on chatgpt's (openai) toes

edit: not sure why I am being downvoted. I am 100% sure the way they structured it was meant to say "we are doing great, but not as great as openAI's work, which we are not trying to compete against". I guarantee there were discussions on how to make it look as to not appear that way.