Codestral Reviews & Product Details

I evaluated Codestral in Mistral Studio through a Rider and Payments-focused engineering agent, using TypeScript coding and code-review prompts. In one task, I specified that a display-name function should trim leading and trailing whitespace, collapse repeated whitespace, preserve capitalization and Unicode characters, and return an empty string for whitespace-only input. Codestral generated the function and Jest tests for the requested cases, using standard JavaScript string methods and a regular expression. The implementation was concise and straightforward to compare against each requirement. I also used follow-up prompts to examine payment-retry logic, idempotency, timeout handling, and possible duplicate processing in a separate illustrative code example. These sessions let me assess both its initial code generation and how it responds when asked to reconsider an earlier solution. Review collected by and hosted on G2.com.

The main limitation I observed was consistency across successive revisions. In the payment-processing example, Codestral identified risks involving duplicate charges, idempotency, and timeout behavior, but some proposed corrections still had retry-termination issues or inconsistencies between the implementation and the suggested tests. I had to inspect the revised code against the original requirements instead of relying on the explanation alone. In the display-name task, the generated implementation met the main requirements, although some of its additional commentary about whitespace edge cases was inaccurate. I would like follow-up responses to preserve earlier requirements, correct the identified defects completely, and keep the tests aligned with the final implementation. Review collected by and hosted on G2.com.

What I like most about Codestral is how useful it is when actually working on code. I just give it my code & ask what's wrong and also to describe different things. It usually gives me something practical to work with. I often use it for writing small functions, error fixing and I speed up my repetitive coding tasks using Codestral. Review collected by and hosted on G2.com.

When it comes to dislikes, I'd probably say the consistency and structure. Sometimes the code given by it looks good at first but doesn't fit in my project structure. I occasionally need to explain the code or give it more context before anything else. It sometimes just overcomplicates the simple solution when I just want a small change... Review collected by and hosted on G2.com.

As a student, the bug-fixing feature has been really useful for my coding assignments. I had a Python code that wasn’t working properly so I highlighted the code, pressed Ctrl+I, and asked it to fix the bug. It found the problem right away and gave me a working solution. It saves me a lot of time compared to going through the code line by line. Review collected by and hosted on G2.com.

As a student, the main issue I have is the response speed. Sometimes I ask it something while working on an assignment, but it takes longer than expected to respond, which can be a bit frustrating when I am trying to finish my work quickly. The setup was also a little confusing at first. I didn’t realize Codestral needs a separate API key from the regular Mistral one, so I had to figure that out before I could actually use it. Review collected by and hosted on G2.com.

Codestral is an LLM by Mistral for coding tasks, and I’m able to use it as one of the models approved by my workplace. I’ve used it to review and evaluate code I need in my day-to-day work, which is mostly (though not limited to) SQL, DAX, and VBA.

The best part of Codestral is how fast it is, and that matters a lot for the way I use coding assistants. When I’m writing code myself, I don’t want an LLM to generate an entire solution end to end; I mainly want help drafting specific pieces. In that situation, quick responses are what count, because I’m typically looking for small snippets that I can refactor and then plug into my existing code.

For these kinds of snippets, since they’re usually low complexity, LLMs generally aren’t wrong, so the speed Codestral offers is genuinely useful. It’s also generally to the point and doesn’t produce overly long answers. Review collected by and hosted on G2.com.

I use coding assistants like Codestral to generate code snippets that I then incorporate into my codebase. I’ve also used Codestral to debug issues in existing code, and most of my requests have been around SQL, DAX, and VBA.

That said, there are situations where Codestral can fall short compared with the competition. I tested some of my SQL prompts across Codestral and another LLM, and for certain prompts Codestral could have performed better. In one instance, I was unsure whether two pieces of code were equivalent. The other LLM stated right at the beginning that they were different, while Codestral said they were the same for a specific use case and then explained the outputs. Only later in its response did it implicitly suggest that the outputs could differ for certain inputs, whereas the other LLM made that point explicitly.

In another test case, I wanted to generate SQL code for some specific use cases and provided three examples. The other LLM recognized a pattern and produced more general code that could work beyond those three examples, but Codestral relied on a SWITCH/CASE-style approach that seemed tailored only to the examples I gave. These are just two specific cases, but they highlight that there’s room for Codestral to improve. Review collected by and hosted on G2.com.

Codestral stands out for its fast, accurate code generation and low-latency completions, making it feel responsive during development. It performs well across multiple programming languages and is especially useful for code completion, debugging, refactoring, test generation, and explaining unfamiliar code. Its support for fill-in-the-middle (FIM) completion makes editing existing code much more natural, and the large context window helps it understand bigger codebases without losing context. Overall, it boosts developer productivity while reducing repetitive coding tasks Review collected by and hosted on G2.com.

While Codestral is fast and capable, it can occasionally generate incorrect or overly confident code for complex business logic, so manual review and testing are still necessary. Its understanding of highly project-specific architectures can be limited without sufficient context, and very large repositories may still require breaking tasks into smaller chunks. Additionally, some advanced features depend on the development environment or IDE integration, and generated code may sometimes need refinement to match coding standards and best practices. Review collected by and hosted on G2.com.

What I like best about Codestral is how reliably it handles code generation without overcomplicating things. It produces clean, readable output, and the responses feel consistent even with tricky prompts. It’s fast, stable, and helps you move through development tasks without friction. Review collected by and hosted on G2.com.

The only thing I dislike is that Codestral can feel a bit too confident with certain outputs. It’s great most of the time, but occasionally it makes assumptions in the code that you didn’t ask for, so you still have to double‑check and clean up those extra pieces. Review collected by and hosted on G2.com.

What I like most about Codestral is how fast and accurate its code completions are. It seems to understand the context of my project really well—whether I’m writing API logic or working through a tricky function, the suggestions feel relevant rather than random. The performance is impressive, too: completions show up almost instantly, which helps me stay in my flow. On their side, the documentation is clear, and getting started took very little time. Review collected by and hosted on G2.com.

Honestly, there’s very little to dislike. If I had to point to anything, I’d love to see a few more editor integrations, even though the main ones I use are already supported and work well. Review collected by and hosted on G2.com.

What I like most is how quickly Codestral provides useful code suggestions. It’s especially good at completing functions, fixing small mistakes, and taking care of repetitive coding tasks. It saves me time, and I don’t have to keep rewriting the same basic code over and over. Review collected by and hosted on G2.com.

Sometimes the suggestions aren’t completely correct, especially with more complex code or when there’s a lot of context. I still have to review what it generates rather than using it blindly. It can also struggle at times with larger, multi-file tasks. Review collected by and hosted on G2.com.

What stands out best is Codestral’s exceptional Fill-in-the-Middle (FIM) performance and its massive 32k context window. Natively optimized to predict and insert missing code right inside existing files, it operates with incredibly low latency. Its ability to ingest large portions of a codebase all at once means it doesn't lose track of multi-file structures, making it a highly reliable co-pilot for complex repository refactoring. Review collected by and hosted on G2.com.

What stands out worst is Codestral’s heavy hardware requirement for local deployment. While it is open-weight, running the model locally with its full 32k context window active demands significant GPU memory and processing power. Additionally, its specific commercial restrictions require paid licensing for enterprise product deployment, making it less accessible for startups wanting to embed it into proprietary software. Review collected by and hosted on G2.com.

N/A Review collected by and hosted on G2.com.

Codestral's code completions and logic generation are noticeably fast compared to standard chat models. I use it while working on college programming assignments in Python and C++. It directly outputs clean functions, algorithmic logic, and correct syntax without adding unnecessary conversational filler or explanations unless requested. Review collected by and hosted on G2.com.