Marie Pellat

Introducing Voxtral Transcribe 2, next-gen speech-to-text models by @MistralAI. State-of-the-art transcription, speaker diarization, sub-200ms real-time latency. Details in 🧵

Mistral is back to their roots All apache 2.0 Strong models, multimodal, multiple types, and MORE BASE MODELS Huge W, smart by them to capitalize on the power vacuum of companies looking for alternates to Chinese models. So happy Chinese companies are pushing change globally.

@nsthorat @dsmilkov @AnthropicAI Seems like a match made in heaven ! Congrats to you both

@sarahookr Congrats 💫💫

I'm starting a new project. Working on what I consider to be the most important problem: building thinking machines that adapt and continuously learn. We have incredibly talent dense founding team + are hiring for engineering, ops, design. Join us: adaptionlabs.ai

We’ve raised €1.7B to accelerate technological progress with AI! This Series C funding round, led by @ASMLcompany, fuels Mistral AI scientific research to keep pushing the frontier of AI to tackle the most critical technological challenges faced by strategic industries.

A win for open science today - 🚢🚢

@spolu Thanks for sharing your thoughts on this, I agree on questioning the motives here. I also liked this essay that touches on the same debate and on the reality of technology taking decades to truly permeate across industries aisnakeoil.com/p/agi-is-not-a…

The current trope on entry-level white-collar jobs being wiped out is interestingly reminiscent of Meta making us believe that we would all work from home in the Metaverse: projection of a few people desires for a technology onto a world view that is seriously disconnected from reality. First, it is important to note that it is quite recommendable for AI leaders to be warning about that. But, the trick of hand here is that, with AI, warning about dangers of the technology and hyping the technology are quasi-indistinguishable and surely have the same effect. So it's important to question the motives. Not saying they are misplaced, just an important dimension to take into account when listening to this discourse which has been on many other labs playbook for quite some time now. Second, even if we were to eradicate entry-level jobs with AI, it's important to realize that this would not happen overnight. Even at current levels it will take at least a decade for this technology to truly permeate the workplace, either through adoption by the largest companies or by competition by AI-first/focused new entrants. Stepping back this is the non compressible time required for the workplace to adapt to this new technology, evolving jobs to take into account this new reality, in some place more rapidly than others, some jobs more aggressively than others, but no wiping involved. We vividly see today in the market that some companies adopt agents very quickly while others have a hard time absorbing them. The resistance being strongly anti-correlated with the evolution capacity of the organization. AI similar to the computer will transform how people work. Yes, some jobs will be transformed so much that they will be hardly recognizable but, similar to the computer, this technology will accelerate society and will likely create better work not less. This is the key: better work, not less. Sells less papers than warning about half of the workforce being wiped but this is the true north star. Computers didn’t replace us, they augmented us, made us more productive and eventually happier in our jobs. Similarly to how it was unclear at some point what would be the human - computer interface we would converge on (the graphical interface eventually) it is still very unclear what will be the AI - human interface we’ll converge on for work. This is our mission to explore that new frontier and transform how works gets done for the better.

We shipped higher level apis for agent orchestration - MCP compatible, all server-side logic deployable privately.

Very happy to release Codestral Embed, our first code embedder. It can use up to 3072 dimensions, ordered by relevance. Using embeddings of 256 dimensions with int8 precision is already sufficient to outperform all existing models.

Always be shipping 🚢 meet the latest edition to our model family Devstral - the best open coding agents model out there. Excited to see what gets built on top of this one !

Meet Devstral, our SOTA open model designed specifically for coding agents and developed with @allhands_ai mistral.ai/news/devstral

Introducing Mistral Medium 3: our new multimodal model offering SOTA performance at 8X lower cost. - A new class of models that balances performance, cost, and deployability. - High performance in coding and function-calling. - Full enterprise capabilities, including hybrid or on-premises/in-VPC deployment, custom post-training, and seamless integration into enterprise tools and systems. Check out our blog to learn more:

Very proud of this cross-institutional collaboration between @Cohere_Labs @stanfordnlp , @MistralAI , @AdaLovelaceInst , @Cambridge_Uni, @Princeton Now accepted to FAccT 2025 🔥 "Is the spotlight on AI amplifying biorisk supported by scientific evidence?

#1 productivity app in France 🚀 Download @MistralAI Le Chat app and give it a try!

Le Chat is fast (1,100 tok/s for flash queries on an updated Mistral Large). Download it at mistral.ai/app/android or mistral.ai/app/ios

No this video is not sped up, genuinely mind blowing And yes this is available to all users right now.

Introducing the all new Le Chat: your ultimate AI sidekick for life and work! Now live on web and mobile!

⚡️⚡️⚡️Le Chat grows up - check it out ⚡️⚡️⚡️ mistral.ai/en/news/all-ne… also if you want speed, we got you youtu.be/_JS-LkBrsk8?si…

Introducing Small 3, our most efficient and versatile model yet! Pre-trained and instructed version, Apache 2.0, 24B, 81% MMLU, 150 tok/s. No synthetic data so great base for anything reasoning - happy building! mistral.ai/news/mistral-s…

The new codestral is tied #1 on the copilot arena! The other models tied for #1 are DeepSeek 2.5 (236B total params, 21B active) and Claude 3.5 Sonnet. They are both much larger than Codestral 25.01.