Mistral AI is moving further into direct competition with the biggest names in artificial intelligence. The Paris-based AI startup has launched Mistral Large, a new flagship large language model, alongside Le Chat, its own beta chat assistant.
The announcement shows how quickly Mistral AI is expanding beyond its early open source positioning. It is now offering a top-tier model through a paid API, testing a consumer-style assistant and adding Microsoft Azure as a distribution channel.
A new flagship model for high-end reasoning
Mistral Large is built as the company’s most advanced large language model. According to the source article, it is designed to rival top-tier models such as GPT-4 and Claude 2 when it comes to reasoning capabilities.
The company says Mistral Large ranks second after GPT-4 on several benchmarks. That claim matters because benchmarks are one way AI companies present model performance to developers and enterprise buyers. Still, the source notes that it is hard to judge real-world performance from benchmark claims alone, because benchmark selection and actual usage can differ.
The model supports English, French, Spanish, German and Italian. By default, Mistral AI supports context windows of 32k tokens, which the source describes as generally more than 20,000 words in English.
That context window is important for users who need a model to process longer prompts, documents or conversations. It defines how much text the system can handle at once, shaping the kinds of workflows developers can build around it.
The paid API looks more like OpenAI’s model
Mistral AI first became known partly for its open source focus. Its first model was released under an open source license with access to model weights. Mistral Large, however, is not being released that way.
Instead, the company is offering Mistral Large through a paid API with usage-based pricing. The source article says it costs $8 per million of input tokens and $24 per million of output tokens to query Mistral Large.
Tokens are the small pieces of text that AI models process. The source gives the example that “TechCrunch” would be split into two tokens, “Tech” and “Crunch,” when handled by an AI model.
For comparison, GPT-4 Turbo has a 128k-token context window and costs $10 per million of input tokens and $30 per million of output tokens. Based on those figures, Mistral Large is currently 20% cheaper than GPT-4 Turbo.
That comparison is useful, but not final. The source notes that pricing in AI changes quickly and companies update prices regularly. For developers, the decision is not only about headline cost; it also depends on model quality, latency, language support and how the model performs in their actual applications.
Le Chat brings Mistral AI to a broader audience
Mistral AI is also launching Le Chat, a chat assistant that works as an alternative to ChatGPT. The service is available in beta, and anyone can sign up at chat.mistral.ai.
Access is free for now. Users can choose between three models:
- Mistral Small
- Mistral Large
- Mistral Next, a prototype model designed to be brief and concise
The company says the beta release may have “quirks.” The source also notes that Le Chat cannot access the web when users interact with it.
That limitation changes what users should expect from the product. Without web access, Le Chat is not positioned in the source as a tool for live browsing or current-information retrieval. It is better understood, based on the article, as a way to try Mistral’s models through a conversational interface.
Mistral AI also plans to launch a paid version of Le Chat for enterprise clients. That version is expected to include central billing and the ability for enterprise clients to define moderation mechanisms.
Microsoft gives Mistral another route to customers
The announcement also includes a partnership with Microsoft. In addition to Mistral AI’s own API platform, Microsoft will provide Mistral models to Azure customers.
On one level, that means Mistral models become another option in Azure’s model catalog. On another, it creates a larger distribution path for Mistral AI, because Azure customers can access the models through a cloud platform they may already use.
The source article also says Mistral AI and Microsoft are holding talks for collaboration opportunities and potentially more. Microsoft made a $16M investment in Mistral AI.
Microsoft is already the main investor in OpenAI’s capped profit subsidiary. At the same time, it has welcomed other AI models onto Azure, including through a partnership with Meta to offer Llama large language models on the platform.
This approach helps Microsoft keep Azure customers inside its product ecosystem while offering more model choice. The source also notes that the strategy might help when it comes to anticompetitive scrutiny.
A fast-moving startup becomes a broader AI platform
Mistral AI was officially incorporated in May 2023. A few weeks later, it raised a $113 million seed round. In December, it closed a $415 million funding round led by Andreessen Horowitz (a16z).
The company was founded by alums from Google’s DeepMind and Meta. Its early reputation centered on open source AI models, but its latest moves show a broader commercial strategy.
With Mistral Large, the company is selling a flagship model through a paid API. With Le Chat, it is testing a direct user-facing assistant. With Microsoft Azure, it is adding a distribution channel that could help the company reach more customers.
The result is a clearer challenge to OpenAI and Anthropic. Mistral AI is no longer only presenting itself as an open source-focused AI lab. It is building the pieces of a larger AI business around models, chat products, enterprise features and cloud distribution.