Researcher

Srijan Mishra

Srijan Mishra is a machine learning researcher whose work at Mistral AI includes Magistral, the company's first reasoning model, and the Voxtral speech models. A machine learning engineer at Uber from 2019 to 2020 and a machine learning scientist at Apple from 2020 to 2024, they joined Mistral AI in November 2024. In 2023, they coauthored the R2 loss paper on low-bit model quantization and compression with Arnav Kundu, Chungkuk Yoo, Minsik Cho and Saurabh Adya. At Mistral, they were among the authors of Magistral in June 2025, the Voxtral speech models in July 2025, the Devstral report in August 2025, and Ministral 3, Voxtral Realtime and Voxtral TTS in 2026. They are an AI research scientist at Mistral AI, working on multimodal LLMs and reinforcement learning.

Works
6
Citations
6
h-index
2

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2019–2024

    Engineer at Uber, scientist at Apple

    They work as a machine learning engineer at Uber from 2019 to 2020, then as a machine learning scientist at Apple from 2020 to 2024.

  2. 2023

    The R2 loss for quantization

    In 2023, they coauthor R2 Loss: Range Restriction Loss for Model Compression and Quantization with Arnav Kundu, Chungkuk Yoo, Minsik Cho and Saurabh Adya. The loss removes outliers from weights during pre-training, which builds models that suit lower-bit quantization and compression.

  3. November 2024

    Joining Mistral AI

    In November 2024, they join Mistral AI as an AI research scientist.

  4. June 2025

    Magistral, a first reasoning model

    In June 2025, they are among the authors of Magistral, Mistral's first reasoning model, built with the company's own scalable reinforcement learning pipeline. Magistral Medium is trained on top of Mistral Medium 3 with reinforcement learning alone, and Magistral Small is open-sourced under Apache 2.0.

  5. July 2025

    The Voxtral speech models

    In July 2025, they are among the authors of Voxtral Mini and Voxtral Small, two multimodal audio chat models that comprehend spoken audio and text documents. Both models are released under the Apache 2.0 license.

  6. August 2025

    Devstral for coding agents

    In August 2025, they coauthor the Devstral technical report on Devstral-Small, a small 24B open source model for code agents.

  7. January 2026–March 2026

    Ministral 3 and Voxtral models

    In January 2026, they are among the authors of Ministral 3, a family of 3B, 8B and 14B dense language models under the Apache 2.0 license. They coauthor Voxtral Realtime, a streaming speech recognition model, in February 2026, and Voxtral TTS, a multilingual text-to-speech model, in March 2026.

  8. As of October 2026

    Research scientist at Mistral AI

    They are an AI research scientist at Mistral AI, working on multimodal LLMs and reinforcement learning in its Research department.

Career

  1. Since 2024Mistral AIAI research scientist
  2. 2020–2024AppleMachine learning scientist
  3. 2019–2020UberMachine learning engineer

Sources and credits

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

  1. 10 Mistral AI research scientists | Instant Expert instant.expert. Checked 20 October 2026.
  2. R2 Loss: Range Restriction Loss for Model Compression and Quantization arxiv.org. Checked 20 October 2026.
  3. Magistral arxiv.org. Checked 20 October 2026.
  4. Voxtral arxiv.org. Checked 20 October 2026.
  5. Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 20 October 2026.
  6. Ministral 3 arxiv.org. Checked 20 October 2026.
  7. Voxtral Realtime arxiv.org. Checked 20 October 2026.
  8. Voxtral TTS arxiv.org. Checked 20 October 2026.
  9. Srijan Mishra - Google Scholar scholar.google.com. Checked 20 October 2026.