Highlights:

  • Le Chat can now process and ingest larger PDF documents, enormous photos, graphs, and equations in order to summarize their contents and evaluate them for insights.
  • Le Chat now has “agentic AI” support, which allows it to carry out intricate, multi-step activities for users.

Mistral AI’s updated Le Chat competes with ChatGPT’s capabilities. France-based generative AI startup is in race with OpenAI with newly launched updates. The company unveiled new chatbot features that claim to outpace those of ChatGPT, and several robust LLMs, including the latest Pixtral Large and updated Mistral Large.

In order to compete with ChatGPT and Anthropic PBC’s Claude, Mistral’s Le Chat, which translates to “the cat” in French, is getting roughly six new capabilities that make it more of a professional assistant.

For example, like ChatGPT and the generative AI search engine Perplexity, Le Chat can now search the web and return results with citations. It also gets a new “Canvas” tool, which is comparable to ChatGPT’s Canvas and allows users to alter content including web pages, data visualizations, and PowerPoint presentations using voice and text commands.

“You can use the canvas feature to create documents, presentations, code, mockups… the list goes on,” the company reported while announcing the new features. “You’re able to modify its contents in place without regenerating responses, version your drafts, and preview your designs.”

Additionally, Le Chat can now process and ingest larger PDF documents, enormous photos, graphs, and equations in order to summarize their contents and evaluate them for insights.

The company claimed that by integrating Black Forest Labs Inc.’s Flux Pro model for image generation, Le Chat is also improving its image generation capabilities. Better more, according to Mistral, it can now facilitate automated processes for activities like expenditure reporting and invoice processing. This means that Le Chat now has “agentic AI” support, which allows it to carry out intricate, multi-step activities for users. ChatGPT does not now have this capability, but it will soon.

For the time being, at least while they are still in beta testing, most of the new features that were revealed recently will be available for free.

Pixtral Large, a multimodal model that can analyze both text and visuals, is the most intriguing of the new LLMs. After the earlier Pixtral 12B model was launched in September, the Pixtral Large becomes the second model in the Pixtral family.

With a substantial 124 billion parameters, the business claims it is far more capable than several of its competitors’ most potent multimodal models, including Anthropic’s Claude 3.5 Sonnet, Google LLC’s Gemini 1.5 Pro, and OpenAI’s GPT-4o, on several important benchmarks. An LLM’s problem-solving skills are roughly measured by parameters, and higher performance is associated with more parameters.

“Pixtral Large is able to understand documents, charts, and natural images,” the company stated. “The model demonstrates frontier-level image understanding.”

According to Mistral’s Head of developer relations, Sophia Yang, Pixtral Large excels in terms of “multilingual optical character recognition, reasoning, chart understanding and more.”

According to Mistral, the capabilities of OpenAI’s GPT-4o are comparable to those of Pixtral Large, which has a context window of 128,000 tokens and can process up to 30 high-resolution photos or a 300-page book at once. You can get it right now on Hugging Face.

The business debuted a recently revised Mistral Large model in addition to Pixtral Large. According to the business, Mistral Large, its flagship text-understanding model, has been improved in long context understanding in its latest version, Mistral Large 24.11, which makes it more suitable for applications like document analysis.

Mistral is one of several promising AI firms vying for market share with Google and OpenAI, which are thought to be leaders in generative AI. A group of former Google DeepMind and Meta Platforms Inc. personnel created it in April 2023, and after obtaining many multi-million dollar funding rounds, it is reportedly valued at USD six billion. About a dozen AI models have been made available thus far, some for study and some for commercial usage.

While the Mistral Small, Medium, and Large can only be accessed through the company’s application programming interface with a license, its most well-known models—Mistral 7B, Mixtral 8x7B, and Mixtral 8x22B—are all open-source and accessible for download through the Hugging Face platform.