AI startup Mistral announced a new AI model focused on coding: Devstral.

Mistral said Devstral was developed in partnership with AI Company All Hands AI, which is publicly available under the Apache 2.0 license, meaning it can be used commercially without restrictions. Mistral claims that Devstral outperforms other open models, such as Google's Gemma 3 27B and China's AI Lab DeepSeek's V3 verification on SWE-Bench, a benchmark for measuring coding skills.

"Devstral (…) is trained to solve real GitHub problems," Mistral wrote in a blog post provided to TechCrunch. "(i)t runs on a code proxy scaffold, such as OpenHANDS or SWE-AGENT, which defines the interface between the model and the test case (…) Devstral is sufficient to run on a single (NVIDIA) RTX 4090 or Mac with 32GB of RAM, making it ideal for on-premises and on-premises and genes."

Devstral arrived as an AI coding assistant and a model that powers them. Just last month, Jetbrains, the company behind a range of popular application development tools, released its first "open" AI model for coding. AI clothing including Google, Windsurf and Openai have also been played publicly in recent months, both public and available, and optimized for programming tasks.

AI models are still difficult to encode quality software – code-generated AI tends to introduce security vulnerabilities and errors due to weaknesses such as the ability to understand programming logic. However, their commitment to increase coding productivity is driving companies and developers to adopt them quickly. A recent poll found that 76% of developers who used or planned to use AI tools during their development last year.

Mistral previously wading into the auxiliary programming space using Codestral (code generation model). But Codestal has not yet obtained permission to allow developers to use models for commercial applications. Its license expressly prohibits “any internal use of employees in the business activities of the company (a)”.

Devstral, which Mistral calls "research preview", can be downloaded from the AI ​​development platform, including embracing faces and exploiting it through Mistral's API. Its price is $0.1 per million input tokens and $0.3 per million output tokens, which is the raw data of the data used by the AI ​​model. (One million tokens equals about 750,000 words, about 163,000 words longer than "war and peace.")

"Efforts to build a larger proxy coding model that will be available in the coming weeks," Mistral said. Devstral is not a small model itself, but on the smaller side, there are 24 billion parameters. (The parameters roughly correspond to the problem-solving skills of the model, and models with more parameters are usually better than models with fewer parameters.)

Founded in 2023, Mistral is a cutting-edge model lab that aims to build a range of AI-powered services including chatbot platforms, LE CHAT and mobile applications. It has been backed by venture capitalists including ordinary catalysts and has raised more than 1.1 billion euros (about $1.24 billion) to date. Mistral's clients include BNP Paribas, Axa and Mirakl.

Devstral is Mistral's third product launch this month. A few weeks ago, Mistral launched the Mistral Medium 3, an effective universal model. Around the same time, the company launched LE Chat Enterprise, a company-centric chatbot service that offers tools like AI "agent" builders and integrates Mistral's models with third-party services like Gmail, Google Drive, and SharePoint.