Vibe is their agentic CLI application, direct competitor to Claude Code etc. I've been using both Claude Code earlier at work and now Mistral Vibe 2.0 for my personal stuff. In my opinion, Mistral...
Vibe is their agentic CLI application, direct competitor to Claude Code etc.
I've been using both Claude Code earlier at work and now Mistral Vibe 2.0 for my personal stuff. In my opinion, Mistral has significantly closed the gap to its US competition, making it a valid EU-based option for a lot of things.
I've been saying this for a while: The LLMs are already pretty great, the tooling still needed to catch up. I think there's still a lot more to be gained, even with current models. I'll definitely...
Mistral has significantly closed the gap to its US competition,
I've been saying this for a while:
The LLMs are already pretty great, the tooling still needed to catch up. I think there's still a lot more to be gained, even with current models.
I haven't tried Vibe yet, so ignore my comment if you are interested in it, but I have given Le Chat from Mistral (it's their ChatGPT alternative) many chances (with and without Pro subscription)...
I haven't tried Vibe yet, so ignore my comment if you are interested in it, but I have given Le Chat from Mistral (it's their ChatGPT alternative) many chances (with and without Pro subscription) in the past and have been thoroughly disappointed on every occasion.
And, as a European, I would love to give up ChatGPT and Gemini, but Mistral just doesn't hold a candle to either of them.
From the top of my head:
Le Chat really isn't good with the few European languages (outside of English) I am fluent in and I would have expected that it would have been extensively trained on EU languages.
The Project feature is extremely sub-par... for example: I create a project to manage my Linux Mint desktop with extensive custom instructions and a knowledge-base.md file with additional details about my setup and take it for a spin. My first test: I try to get it to help me with 1Password which is throwing errors sometimes. Le Chat tells me to open Finder (on a Linux Mint desktop) and go to the Applications folder, as if I was using a Mac. That's with custom instructions mentioning Linux and Linux Mint multiple times and an extensive knowledge-base.md file.
I'll continue following Mistral and trying it out, but it needs to get much better if it wants to compete with the big leagues.
I imagine this has to do with the corpus of text available in different languages. English has copious volumes of text on the internet. I'm not sure how many languages compete. Maybe Chinese?
Le Chat really isn't good with the few European languages (outside of English) I am fluent in and I would have expected that it would have been extensively trained on EU languages.
I imagine this has to do with the corpus of text available in different languages. English has copious volumes of text on the internet. I'm not sure how many languages compete. Maybe Chinese?
Yeah, English is the default language of AI because of the available amount of text. My point was rather that ChatGPT and Gemini cover other languages much better to the point that it's not...
Yeah, English is the default language of AI because of the available amount of text. My point was rather that ChatGPT and Gemini cover other languages much better to the point that it's not possible to use them with Mistral.
I suppose it depends on the languages the person was looking into. I would expect an LLM do to pretty well with major European languages like German, French, Spanish, and Russian, because there is...
I suppose it depends on the languages the person was looking into. I would expect an LLM do to pretty well with major European languages like German, French, Spanish, and Russian, because there is also a lot of content in those languages (albeit not as much as in English). I would have much more moderate expectations for a smaller language like Icelandic or Bulgarian. That said, it seems like the person you were replying to was probably comparing its performance in the relevant languages to that of other LLMs.
I have same and slightly different experiences. I've used Mistral to quite successfully to learn French, but perhaps it's not so surprising that an LLM made by a french company is good at French....
I have same and slightly different experiences. I've used Mistral to quite successfully to learn French, but perhaps it's not so surprising that an LLM made by a french company is good at French.
Yes, the general quality of the LLM is not quite as good, but with Vibe 2.0 (with its devstral model) the gap has closed somewhat. I wouldn't dare to use it on employer projects yet though, but for my personal coding projects (where the stakes aren't high at all) it seems to be fine. It knows how to split tasks into smaller todos, which it's pretty good at implementing.
I have the pro subscription of Mistral, dunno if that matters.
I've tested GitHub Copilot with GPT-5 and Claude Sonnet extensively at work and can say that they are really good. They do require very specific instructions, but I like it that way. I know what I...
I've tested GitHub Copilot with GPT-5 and Claude Sonnet extensively at work and can say that they are really good. They do require very specific instructions, but I like it that way. I know what I want to achieve, I'm just too lazy to write it out. :)
In that sense, I'll have to give Vibe 2.0 a fair chance with some private projects lying around to see how it holds up.
Vibe is their agentic CLI application, direct competitor to Claude Code etc.
I've been using both Claude Code earlier at work and now Mistral Vibe 2.0 for my personal stuff. In my opinion, Mistral has significantly closed the gap to its US competition, making it a valid EU-based option for a lot of things.
I've been saying this for a while:
The LLMs are already pretty great, the tooling still needed to catch up. I think there's still a lot more to be gained, even with current models.
I'll definitely try out mistral yet
I haven't tried Vibe yet, so ignore my comment if you are interested in it, but I have given Le Chat from Mistral (it's their ChatGPT alternative) many chances (with and without Pro subscription) in the past and have been thoroughly disappointed on every occasion.
And, as a European, I would love to give up ChatGPT and Gemini, but Mistral just doesn't hold a candle to either of them.
From the top of my head:
I'll continue following Mistral and trying it out, but it needs to get much better if it wants to compete with the big leagues.
I imagine this has to do with the corpus of text available in different languages. English has copious volumes of text on the internet. I'm not sure how many languages compete. Maybe Chinese?
Yeah, English is the default language of AI because of the available amount of text. My point was rather that ChatGPT and Gemini cover other languages much better to the point that it's not possible to use them with Mistral.
I suppose it depends on the languages the person was looking into. I would expect an LLM do to pretty well with major European languages like German, French, Spanish, and Russian, because there is also a lot of content in those languages (albeit not as much as in English). I would have much more moderate expectations for a smaller language like Icelandic or Bulgarian. That said, it seems like the person you were replying to was probably comparing its performance in the relevant languages to that of other LLMs.
I have same and slightly different experiences. I've used Mistral to quite successfully to learn French, but perhaps it's not so surprising that an LLM made by a french company is good at French.
Yes, the general quality of the LLM is not quite as good, but with Vibe 2.0 (with its devstral model) the gap has closed somewhat. I wouldn't dare to use it on employer projects yet though, but for my personal coding projects (where the stakes aren't high at all) it seems to be fine. It knows how to split tasks into smaller todos, which it's pretty good at implementing.
I have the pro subscription of Mistral, dunno if that matters.
I've tested GitHub Copilot with GPT-5 and Claude Sonnet extensively at work and can say that they are really good. They do require very specific instructions, but I like it that way. I know what I want to achieve, I'm just too lazy to write it out. :)
In that sense, I'll have to give Vibe 2.0 a fair chance with some private projects lying around to see how it holds up.