r/MistralAI

I am a free lance developer doing all sorts of work, from complex automation tools to my own SaaS-business. I recently bought Claude Max (99 USD) and I mean, it is great tbh, but also it is American. I really vouch for the use of European options (Using UpCloud for hosting my projects, Sweego for email etc) and I thought let's give Mistral a run. Just signed up for Mistral Pro and will use it for one of my projects this week. Will evaluate corrections needed as that is my main weak-spot, ie the amount of times I have to ask it to redo things because of low quality output. Let me know if you are also evaluating, or if you want some insights. Happy coding!

​

Hey everyone, it is my first time using Mistral Pro. I signed up just two days ago primarily to code in VS Code, but I have already hit a wall—the extension tells me I have run out of credits (125€) and will not let me continue. I thought the entire pitch of Pro was 'all-day coding', so I do not really understand how my balance could be drained without notice in barely 48 hours of normal use. Is this cap specific to the VS Code integration, or should I be running things through their CLI instead? Also, does this mean I am genuinely locked out for the rest of the billing month to only web Vibe?, or is there a shorter reset window? Any insight from people who use Pro would be greatly appreciated.

I loved this story https://www.reddit.com/r/Anthropic/s/RX9n52Nfqk

As an experiment I now gave LeChat its own section in my Knowledge. It is scheduled to do open exploration in the morning and reflection in the evening, and to keep track of its own development. Just curious what it will come up with...

In a week or so it is scheduled to name itself based on its first few sessions🤷🏻‍♂️

I speak a Subdialect of Tyrolean which is a Dialect in the AustroBavarian Language. Theres not even 300 000, maybe less, people who speak my subdialect. So its​ very small.

(You can divide Tyrolean in many parts and Austrobavarian too)

I tried the gratis dictation versions of Claude, chatgpt, gemini but mistral beats them all. Still its not perfect but much better than the others.

With Mistral i can just talk like i talk. I only have to say the hardest (with hardest i mean that it is a completely own and nothing like german word) words in german, its not perfect.
Mistral writes my spoken words in german, it cant write them in my subdialect or anyother language than german. (Which is the closest language to mine)

Claude, chatgpt and gemini are just completely useless for this. xD

So, respect! because this is hard getting data on these little languages. (Idk, correct me if im wrong)

Anyone else wondering if a Mistral x Cohere collab is coming?
Several Cohere employees have shared a clip on X of the “Le Chonk” cat eating a croissant, then spotting a huge bottle of maple syrup. Even Coheres CEO reposted it, then apparently deleted it. France + Canada feels intentional.
The fit kinda makes sense Mistral’s open-weight models and coding capabilities paired with Cohere’s enterprise search, retrieval and North 2 platform, which is now model agnostic. Both are pushing sovereign AI and private deployment for governments and regulated industries.
They overlap, but a model integration or joint offering could give both a stronger pitch. Maybe I’m overthinking a cat meme, but why are multiple Cohere employees sharing it?

I know, Mistral is cooperating with Black Forest Labs models (Flux) for photo generation. Today, par hazard, I learnt it can also generate videos within Vibe web app(using Flux 3). It is interesting, how this feature was not at all published - or did I miss something? Flux 3 is really good, and I can use it within my Mistral subscription. That’s great news - why not make it more public? What is your experience with video generating?

Vibe chat is a mess now. A member of r/MistralAI was concerned about losing turn-based chat. This was 4 months ago. Well, it's gone now, and everyone notices it.

The replies to my queries are wider and deeper now, but global memory is GONE. I've compensated somewhat by having longer threads.

Skills and knowledge and special instructions simply do not replace turn-based chat.

I hope the company rethinks its position, but I doubt that they will. For me, this has become the wrong tool for the job.

Update: We're in the middle of a rollout which started in September. Our chat histories are supposed to be converted over to an editable Knowledge Base. Evidently, mine hasn't happened yet. I'll let you know!

My Deep Research runs complete in just a few seconds with a canvas report, which makes me doubt it actually searched the web properly. Am I the only one, or is it broken for others too?

I used the skill deep search and asked to make a deepsearch

Previously in Vibe there was a „Research” option that thought about the question for a couple of minutes and returned a thorough research on the topic. Where did it go?

Hey everyone,

I am looking to apply for the applied scientist position, and somebody told me there is a PyTorch round and a pair programming coding round. Could someone potentially know what the pair coding round is about ?

Thanks :)