Skip to main content

This open-source alternative to ChatGPT just got serious

The beta Canvas feature on Mistral
Andrew Tarantola / Mistral

French AI startup Mistral announced Monday that it is incorporating a half-dozen new features and capabilities into its free generative AI work assistant, dubbed le Chat (French for “the cat”), that will put the open-source chatbot on par with leading frontier models from OpenAI and Anthropic.

Le Chat can now search the web and provide cited sources, similar to what Perplexity and SearchGPT both offer. Mistral’s chatbot now also offers a Canvas feature akin to Claude’s Artifacts where users can modify and edit content and code. What’s more, le Chat can now generate images thanks to an integration with Black Forest Labs’ Flux Pro, the same image generator that powers Grok-2’s capabilities.

Canvas

Beyond that, le Chat can now analyze and summarize the contents of large PDF files, including any graphs and equations. The company is even rolling out AI “agents,” beating OpenAI to the punch, which can automate workflows for repetitive tasks and be shared among a company’s employees.

Recommended Videos

Many of these features are only made possible by Mistral’s new model, the 128B-parameter Pixtral Large. Built atop Mistral Large 2, this frontier model offers a 128k prompt window and state-of-the-art performance on a number of industry benchmarks, including MathVista, DocVQA, and VQAv2.

“Particularly, Pixtral Large is able to understand documents, charts, and natural images,” the company wrote in its announcement post. Mistral is also debuting a new version of its flagship Mistral Large model. Version 24.11 reportedly offers improved performance in tasks that require larger amounts of context, such as document analysis.

“At Mistral, our approach to AI is different — we’re not chasing artificial general intelligence at all costs; our mission is to instead place frontier AI in your hands, so you get to decide what to do with advanced AI capabilities,” the company wrote. “This approach has allowed us to be quite frugal with our capital, while consistently delivering frontier capabilities at affordable price points.” As such, all of the features debuting today are currently free to use for all le Chat users.

Mistral was founded in April 2023 by former Meta Platforms and Google DeepMind employees. The company is currently valued at $2 billion and offers nearly a dozen different AI models for commercial and research use.

Three of the models — Mistral 7B, Mixtral 8x7B, and Mixtral 8x22B — are open source and available to the public through Hugging Face. The Mistral Small, Medium, and Large models, conversely, are closed-source and only available through the Mistral API. Users can access both Large 24.11 and Pixtral Large with either a commercial use license or a more restrictive research license from Mistral.

Andrew Tarantola
Former Digital Trends Contributor
Andrew Tarantola is a journalist with more than a decade reporting on emerging technologies ranging from robotics and machine…
DeepSeek invites users behind the curtain of its open source AI code
Phone running Deepseek on a laptop keyboard.

The Chinese startup, DeepSeek plans to become even more transparent about the technology behind its open-source AI models, such as its R1 reasoning model.

The company detailed in a post on X on Friday that it will make several code repositories available to the public, starting next week. This will give developers and researchers a deeper understanding of the nuances of the key parts of DeepSeek’s code. It is an especially bold move for a tech company. However, bold moves are already par for the course for DeepSeek, which entered the AI space as an industry disrupter. It has especially stood out because its models have performed as well, if not better than many of the top AI brands in the industry, such as OpenAI and Meta– that use proprietary technologies.

Read more
OpenAI’s Operator agent is coming to eight more countries
ChatGPT and OpenAI logos.

Following  its U.S. debut in January, OpenAI's Operator AI agent will soon be expanding to eight new nations, the company announced on Friday.

"Operator is now rolling out to Pro users in Australia, Brazil, Canada, India, Japan, Singapore, South Korea, the UK, and most places ChatGPT is available," the OpenAI team wrote in a post to X. The company is "still working on making Operator available in the EU, Switzerland, Norway, Liechtenstein & Iceland," but has not clarified release timing for those additional countries. As with the American release, users in the expanded nation list will still have to pay for OpenAI's $200 per month Pro tier subscription in order to access the AI agent.

Read more
With 400 million users, OpenAI maintains lead in competitive AI landscape
OpenAI's new typeface OpenAI Sans

Competition in the AI industry remains tough, and OpenAI has proven that it is not taking any coming challenges lightly. The generative AI brand announced Thursday that it services 400 million weekly active users as of February, a 33% increase in less than three months.

OpenAI chief operating officer, Brad Lightcap confirmed the latest user statistics to CNBC, indicating that the figures had not been previously reported. The numbers have quickly risen from previously confirmed stats of 300 million weekly users in December.

Read more