Mathilde Guillaumin

Timeline

1. 2020

   A Vamp2 sleep mutant

   At the Sleep and Circadian Neuroscience Institute in Oxford's Nuffield Department of Clinical Neurosciences, supported by a Clarendon Scholarship and a BBSRC DTP grant, she is co-first author with Gareth T. Banks of a Science Advances paper published in August 2020. Using forward genetics and in vivo electrophysiology, the team identifies a recessive mouse mutant, caused by an Ile102Asn substitution in the synaptic vesicle protein VAMP2, with a reduced tendency to move between wake and sleep and a marked deficit in starting REM sleep.

2. 2023

   Hedonic eating at ETH Zurich

   At the Institute for Neuroscience in ETH Zurich's Department of Health Sciences and Technology, she is first author of a 2023 Molecular Psychiatry study with Daria Peleg-Raibstein and Denis Burdakov. Using cell-specific recording and optogenetics in mice, the study shows that D1 and D2 cells of the nucleus accumbens encode different aspects of food liking and wanting. She carries out the investigation and co-writes the first draft with Peleg-Raibstein, who supervises the work.

3. 2023–2025

   Postdoc in Helen Barron's group

   From 2023 to 2025, she is a postdoctoral neuroscientist in Associate Professor Helen Barron's group at the University of Oxford's Brain Network Dynamics Unit.

4. January 2024

   The Somnotate sleep classifier

   In January 2024, she is among the co-authors of Somnotate in PLOS Computational Biology, contributing data curation. The probabilistic classifier combines linear discriminant analysis with a hidden Markov model to annotate sleep stages in mouse recordings, exceeds human experts in accuracy and reports the certainty of each annotation.

5. June 2025

   Magistral, Mistral's first reasoning model

   In June 2025, she is among the co-authors of Magistral, Mistral's first reasoning model, built with the company's own scalable reinforcement learning pipeline. The team trains Magistral Medium for reasoning on top of Mistral Medium 3 with reinforcement learning alone and releases Magistral Small under the Apache 2.0 license.

6. July 2025

   Voxtral audio chat models

   In July 2025, she co-authors Voxtral, which presents Voxtral Mini and Voxtral Small, two multimodal audio chat models that understand both spoken audio and text documents. A 32K context window lets the models handle audio files of up to 40 minutes, and both are released under the Apache 2.0 license.

7. January–March 2026

   Ministral 3 and new Voxtral models

   In January 2026, she co-authors Ministral 3, a family of parameter-efficient dense language models in 3B, 8B and 14B sizes derived through Cascade Distillation. Her co-authored papers in February and March 2026 add Voxtral Realtime, a streaming speech recognition model with sub-second latency, and Voxtral TTS, a multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio.

8. Since 2025

   Research at Mistral AI

   She moves from Oxford to Mistral AI in 2025. In July 2026, she co-authors Robostral Navigate, an 8B vision-language model that guides wheeled, legged and aerial robots from monocular camera images, and Shieldstral, a 3B-parameter multimodal safety classifier.