Mistral’s new Codestral code completion model races up third-party charts

Mistral has updated its open-source coding model Codestral — which is proving popular among coders — extending the competition for coding-focused models targeted to developers. 

In a blog post, the company said it has upgraded the model with more efficient architecture to create Codestral 25.01, a model Mistral promises will be the “clear leader for coding in its weight class” and twice as fast as the previous version. 

Like the original Codestral, Codestral 25.01 is optimized for low-latency, high-frequency actions and supports code correction, test generation and fill-in-the-middle tasks. The company said it could be helpful for enterprises with more data and model residency use cases. 

Benchmark tests showed Codestral 25.01 performed better in tests coding in Python and scored 86.6% in the HumanEval test. It beat the previous version of Codestral, Codellama 70B Instruct and DeepSeek Coder 33B instruct. 

This version of Codestral will be available to developers who are part of Mistral’s IDE plugin partners. Users can deploy Codestral 25.01 locally through the code assistant Continue. They can also access the model’s API through Mistral’s la Plateforme and Google Vertex AI. The model is available in preview on Azure AI Foundry and will be on Amazon Bedrock soon. 

More and more coding models

Mistral released Codestral in May last year as its first code-focused model. The 22B parameter model could code in 80 different languages and outperformed other code-centric models. Since then, Mistral released Codestral-Mamba, a code generation model built on top of the Mamba architecture that can generate longer code strings and handle more inputs. 

And, it seems there’s already a lot of interest in Codestral 25.01. Just a few hours after Mistral made its announcement, the model is already racing up the leaderboards on Copilot Arena. 

Writing code was one of the earliest features of foundation models, even for more general-purpose models like OpenAI’s o3 and Anthropic’s Claude. However, in the past year, coding-specific models have improved, and often outperform larger models. 

In the past year alone, there have been several coding-specific models made available to developers. Alibaba released Qwen2.5-Coder in November. China’s DeepSeek Coder became the first model to beat GPT-4 Turbo in June. Microsoft also unveiled GRIN-MoE, a mixture of experts (MOE)-based model that can code and solve math problems. 

No one has solved the eternal debate of choosing a general-purpose model that learns everything or a focused model that only knows how to code. Some developers prefer the breadth of options they find in a model like Claude, but the proliferation of coding models shows a demand for specificity. Since Codestral is trained on coding data, it will, of course, be better at coding tasks instead rather than writing emails. 

Daily insights on business use cases with VB Daily

If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI.

Read our Privacy Policy

Thanks for subscribing. Check out more VB newsletters here.

An error occured.