French AI Startup Mistral is releasing a new AI model, Mistral Medium 3, which focuses on efficiency without compromising on performance.
The price of $ 0.40 per million input tokens and $ 2 per million per million output token in Mistral’s API, Mistral Medium 3 “AT or above performs” 90% anthropic’s Costlier Cloud Sonnet 3.7 model “on the” benchmark across the board “on the 3.7 model. It also crosses recent open models, including the Meta’s Lama 4 MAVIRIC and Kohere’s Command A, popular AI performance evaluation.
Tokens work with the raw bits of the data model, with a million tokens with approximately 750,000 words (approximately 163,000 words are longer than “war and peace”).
In a blog post sent to Techcrunch, Mistral said, “Mistral Medium 3 (…) can be deployed on any cloud, including four GPU and above self-hosted environment.” “On pricing, the model defeated leaders such as Deepsek V3 in both API and Self-Diploid system.”
Mistral, established in 2023, is a frontier model lab, aimed at building a series of AI-operated services, including a chatbot platform, Le chat and mobile app. It is supported by the VC, including the general catalyst, and has increased by more than € 1.1 billion (about $ 1.24 billion) to date. Mistral customers include BNP Paribas, AXA and Mirakl.
According to Mistral, Mistral Medium 3 is best for coding and stem tasks, and multimodal understands Excel. The company says that the model has been tested for cases such as customer service, workflow automation, and complex dataset analysis to customers in financial services, energy and healthcare.
In addition to Mistral’s API, where enterprise customers can work with Mistral to fix the IT, Mistral Medium 3 is available on the Amazon’s Sammekar platform starting from Wednesday. It will soon come in other hosts, including Microsoft’s Azure AI Foundry and Google’s vertex AI platforms, the company said.
Techcrunch event
Burkeley, CA
,
5 June
book now
The launch of Mistral Medium 3 is on the heels of Mistral Small 3.1 of Mistral in March. In its blog post, the company teased the release of a very large model in the coming weeks.
Mistral also launched the Le Chat Enterprise, a corporate-centered chatbot service on Wednesday, providing equipment such as the AI ”agent” builder, and the model of Mistral integrates the model with third-party services such as Gmail, Google Drive and Sharepoint. Le Chat Enterprise rolled out in private preview earlier this year, but today it has a general availability.
Le Chat Enterprise will soon support the MCP, anthropic standard to connect AI assistants to the system and software where the data remains. Other major AI model providers, including Google and Openai, announced that they would adopt MCP earlier this year.
Updated 7:48 AM Pacific: A pricing mistake fixed – Mistral’s blog post was a typo in the version of Techcrunch. We regret error.
