I found Mistral's agentic coding models to be fast and good, but not always accurate. I think the reason for that is that most IDEs aren't properly setup to use their models. But I haven't tried this new Codestral version.
They're such idiots, they haven't published Mistral Code on OpenVSX. LMAO.
I wanted to recommend to a group of friends to try it, but I can't because they use VSCodium.
It certainly makes them idiots, from the perspective of artificially restricting their audience; AFAIK it's very simple to publish on OpenVSX, there are no extension changes needed. Most devs I know use VSCodium, not VSCode.
And how many devs do you know? What percentage of all the devs in the world does that represent?
If you know a million devs who mostly use VSCodium instead of VSCode, that might be significant. But I'm guessing you just jump to conclusions about almost everything based on your own bubble and lack of critical thinking skills. Then you can just call other people idiots while patting yourself on the back for being so smart.
I think counter to your experience I don’t know a single developer who uses Codium. While I get the frustration, it may come soon but not immediately.
It's not an issue for me, I have enough coding assistants. It's a issue for them, since they are restricting their audience artificially, since it requires minimal effort to publish on OpenVSX.
anyone has experience how Codestral compares to other models?
I found Mistral's agentic coding models to be fast and good, but not always accurate. I think the reason for that is that most IDEs aren't properly setup to use their models. But I haven't tried this new Codestral version.
On livecodebench it's picking up 26%
phi4 is 23%
deepseek r1 qwen distilled 32b is 27%
llama 3.3 70b is 29% same with llama 4 scout
gpt 4o is 31%
gpt 41 is 45%
qwen3 32b reasoning 55%!! Expecting qwen3 coder 30b to be around here?
kimi k2 55%
claude 4 around 60%
qwen3 coder 480b 58%
nemotron 49b 74%!!
glm 4.5 358b 74%
exaone 4 32b reasoning 74%!!
deepseek r1 685b 75%
grok4, o4mini, gemini2.5pro, 80%
Devstral was trained specifically to follow the "agentic" or "tool use" pattern.
Where it does each step. Read the files, then edit in text editor get it right, then write to the file.
Codestral is pre "agentic" the here's the code you copy and paste it.
Devstral > codestral.
They're such idiots, they haven't published Mistral Code on OpenVSX. LMAO. I wanted to recommend to a group of friends to try it, but I can't because they use VSCodium.
And that makes them "idiots"? Is it really necessary to resort to insults because they don't immediately support your edge case?
It certainly makes them idiots, from the perspective of artificially restricting their audience; AFAIK it's very simple to publish on OpenVSX, there are no extension changes needed. Most devs I know use VSCodium, not VSCode.
If I were their investor I would not be happy.
> Most devs I know use VSCodium, not VSCode.
And how many devs do you know? What percentage of all the devs in the world does that represent?
If you know a million devs who mostly use VSCodium instead of VSCode, that might be significant. But I'm guessing you just jump to conclusions about almost everything based on your own bubble and lack of critical thinking skills. Then you can just call other people idiots while patting yourself on the back for being so smart.
I think counter to your experience I don’t know a single developer who uses Codium. While I get the frustration, it may come soon but not immediately.
Seems like a self-imposed issue to me.
It's not an issue for me, I have enough coding assistants. It's a issue for them, since they are restricting their audience artificially, since it requires minimal effort to publish on OpenVSX.
They might be "restricting their audience artificially" by like a dozen people.
> I have enough coding assistants
That's a weird and irrelevant flex. Actually, all of your points and insults are weird and irrelevant.