Khyathi Raghavi Chandu

prof_pic.jpg

I am an AI Research Scientist at Mistral AI, based in Palo Alto. My research centers on developing and training large-scale models—ranging from 3B, 24B to 123B parameters—with a focus on two key directions:

🎧 Multimodality with Voxtral: a family of open-source audio–language models.

🧠 Reasoning with Magistral: a line of large-scale reasoning models.

A core part of my efforts is advancing open-source research and releases to make these capabilities broadly accessible to the community.

Previously, I was part of the Mosaic team at the Allen Institute for AI, working in Yejin Choi’s group, and also worked as a Research Scientist at Meta AI. I earned my Ph.D. from Carnegie Mellon University at the Language Technologies Institute, where my research centered on grounded multimodal long-form generation, advised by Alan W. Black, Eric Nyberg, and Yonatan Bisk.

Get in touch if you want to chat, I’m open to collaborations or brainstorming!

I am fortunate to have been selected as a Rising Star by UC Berkeley. I am very passionate about DEI and co-chaired Academic Inclusion sub-committee multiple years at ACL 2020, ACL 2021, NAACL 2022 and co-organized Widening NLP in 2020 and 2021. I am very active in participating in community activities, organized several *CL workshops at conferences and served as Area Chair at COLM 2024, ACL ARR (Feb, Apr, June) 2024, and AACL 2023. I am fortunate to have worked with amazing researchers in my past internships at Google, Apple and Meta.

News

Jul 15, 2025 Announced Voxtral (Small and Mini), Mistral AI’s first audio model, frontier class audio model with open-weights, handles 32k context length and handles audio upto 40 minutes with Apache 2.0 license.
Jun 10, 2025 Announced Magistral Medium and Small, Mistral AI’s first reasoning model.
Aug 14, 2024 [2 Best Paper Awards] @ ACL 2024 for OLMo and Dolma
Jul 2, 2024 New preprint alert! Certainly Uncertain improves model’s ability to say IDK and introduces a new confidence-weighted metric.
Jun 17, 2024 New paper alert! DatComp-LM is out. A large effort from over 20 institutes!
May 16, 2024 5 papers accepted at ACL, say Hi if you are there!
Mar 1, 2024 2 papers accepted at ICLR 2024!
Dec 4, 2023 Organizing workshops on WNU at EMNLP 2024. Stay tuned for awesome keynote talks and presentations!
Aug 26, 2023 Invited talk @ Netflix on “Decoding Uncertainty in Knowledge Quadrant””
intro