way worse then qwen3 coder or devstral for coding not recommended :(
Tried the official magistral_small_2509 and it is way worse then qwen3 coder or devstral.
Code compare gives these results:
"Why this matters:
The first version's directory handling had a critical flaw: it would attempt to create parent directories recursively but didn't handle failures properly. This could lead to the application appearing to hang or behave unpredictably when trying to create output folders.
The second version fixes these issues with clean, standard Windows API usage and proper error handling that follows Microsoft's recommended patterns for directory operations.
Conclusion:
folder create bug fix2.txt is clearly superior in robustness and quality. It addresses critical bugs present in the first version while improving user experience through better error messages and more reliable operation. The code also aligns with standard Windows programming practices, making it easier to maintain and extend.
The second version demonstrates professional software engineering practices that would prevent common issues users might encounter when trying to process files into non-existent output directories - a very real scenario for the application's target use case."
never mind, this model is even worse than 2507
I noticed that some regions near the top of the tensor slice look "flat" several tensor slices have nearly identical values
https://www.reddit.com/r/MistralAI/comments/1o0kvpe/weird_repeating_values_in_magistralsmall2509_gguf/
(EDIT: It's look normal)
So mistral released faulty model, great for them. Great advertising for their commercial versions.