PeopleofSuperintelligence

Researcher

Joachim Studnia

Joachim Studnia is an AI scientist at Mistral AI who co-authored Magistral, Mistral's first reasoning model, along with the reports on the Voxtral audio models and the Devstral coding-agent model. After entering École polytechnique with the X2019 class, he studied at Stanford, where he built a multitask BERT model for the CS224N course and completed a master's degree in 2024. In 2023, he was first author of a NeurIPS workshop paper on defending large language models against adversarial attacks. His model work runs from the multimodal Pixtral 12B in 2024 to the Ministral 3 family and the Voxtral Realtime speech recognizer in 2026. In July 2026, he co-authored Shieldstral, a multimodal safety classifier, and Robostral Navigate, a vision-language model for robot navigation.

Badges

Timeline

  1. 2019–2025

    École polytechnique, class of X2019

    He enters École polytechnique in 2019 through the MP computer-science track of its entrance exam, joining the X2019 class. He receives the polytechnique diploma with the X2019 class in March 2025.

  2. 2023–2024

    A master's at Stanford

    In Winter 2023, he builds a multitask BERT model as his final project for Stanford's CS224N course. He completes a master's degree at Stanford in 2024.

  3. 2023

    Adversarial defenses for LLMs

    In 2023, he is first author of "Evaluating Adversarial Defense in the Era of Large Language Models," a spotlight paper at the NeurIPS 2023 R0-FoMo workshop, written with Simiao Zuo, Xiaodong Liu, Qiang Lou, Jian Jiao and Denis Charles. The paper proposes prompting, typo-correction and fine-tuning defenses that raise LLM robustness by up to 20%.

  4. October 2024

    Pixtral 12B

    In October 2024, he co-authors Pixtral 12B, a 12-billion-parameter multimodal language model that understands natural images and documents. Its new vision encoder, trained from scratch, takes in images at their natural resolution and aspect ratio, and the model is released under the Apache 2.0 license.

  5. June 2025

    Magistral, a reasoning model

    In June 2025, he co-authors Magistral, Mistral's first reasoning model, built on the company's own scalable reinforcement learning pipeline. The work yields Magistral Medium, trained for reasoning on top of Mistral Medium 3 with RL alone, and the open-source Magistral Small.

  6. July–August 2025

    Voxtral and Devstral

    In July 2025, he co-authors Voxtral, which introduces Voxtral Mini and Voxtral Small, two multimodal audio chat models that understand spoken audio and text. In August 2025, he co-authors the technical report on Devstral-Small, a 24B open-source model for coding agents.

  7. January–February 2026

    Ministral 3 and Voxtral Realtime

    In January 2026, he co-authors Ministral 3, a family of dense language models in 3B, 8B and 14B sizes, derived through Cascade Distillation and equipped with image understanding. In February 2026, he co-authors Voxtral Realtime, a streaming speech recognition model that matches offline transcription quality at sub-second latency.

  8. As of October 2026

    AI scientist at Mistral AI

    He is an AI scientist at Mistral AI. In July 2026, he co-authors Shieldstral, a 3B-parameter multimodal safety classifier, and Robostral Navigate, an 8B vision-language model for robot navigation that works from a stream of monocular RGB images.

Career

  1. NowMistral AIAI scientist

Education

  1. 2025École PolytechniquePolytechnique diploma
  2. 2024Stanford UniversityMaster's degree

From the biography.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Joachim Studnia. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  2. 16% of AI career moves in the 2020s went from a university to a companyIncludes Joachim Studnia. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  3. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Mistral AI. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  4. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Mistral AI. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Joachim Studnia yet. Send us a free-licensed photo.

  1. Joachim Studnia - JORFSearch jorfsearch.steinertriples.ch. Checked 20 October 2026.
  2. joa-stdn (Joachim Studnia) · GitHub github.com. Checked 20 October 2026.
  3. NeurIPS Evaluating Adversarial Defense in the Era of Large Language Models neurips.cc. Checked 20 October 2026.
  4. Evaluating Adversarial Defense in the Era of Large Language Models | ML Anthology mlanthology.org. Checked 20 October 2026.
  5. [2410.07073] Pixtral 12B arxiv.org. Checked 20 October 2026.
  6. [2506.10910] Magistral arxiv.org. Checked 20 October 2026.
  7. [2507.13264] Voxtral arxiv.org. Checked 20 October 2026.
  8. [2509.25193] Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 20 October 2026.
  9. [2601.08584] Ministral 3 arxiv.org. Checked 20 October 2026.
  10. [2602.11298] Voxtral Realtime arxiv.org. Checked 20 October 2026.
  11. [2607.25857] Shieldstral arxiv.org. Checked 20 October 2026.
  12. [2607.20785] Robostral Navigate arxiv.org. Checked 20 October 2026.