PeopleofSuperintelligence

Researcher

Soham Ghosh

Soham Ghosh is a research scientist at Mistral AI whose work spans multimodal models, computer vision, reasoning, and speech. He earned engineering and management degrees at Nanyang Technological University, then completed an M.S. in computational data science at Carnegie Mellon while working on video-language grounding and meta reinforcement learning. In 2019, he coauthored ExCL, a method for locating video clips from language descriptions. His earlier research at Google DeepMind focused on audio-visual representation learning; he coauthored TIPS and Pixtral 12B in 2024 and Magistral in 2025. His Mistral work includes Voxtral audio models and Voxtral TTS, a multilingual speech-generation system.

Works
43
Citations
355
h-index
9

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2012–2016

    Engineering and management degrees

    He earns a B.Eng. and M.Sc. in Engineering Science (Computer Science) and Technology Management at Nanyang Technological University from 2012 to 2016.

  2. 2017–2018

    A master's at Carnegie Mellon

    From 2017 to 2018, he studies Computational Data Science (Analytics) for an M.S. at Carnegie Mellon University's School of Computer Science. During the degree, he works with Alexander Hauptmann on video-language grounding and Ruslan Salakhutdinov on meta reinforcement learning.

  3. 2019

    Extractive clip localization

    In 2019, he coauthors ExCL: Extractive Clip Localization Using Natural Language Descriptions. The method predicts the start and end frames for a video clip from a text description through cross-modal interactions between text and video.

  4. 2024

    TIPS image-text pretraining

    His earlier research at Google DeepMind focuses on audio-visual representation learning with language supervision. In 2024, he coauthors TIPS, a general-purpose image-text model for dense and global vision tasks.

  5. October 2024

    Pixtral 12B

    In October 2024, he coauthors Pixtral 12B, a 12-billion-parameter multimodal language model that understands natural images and documents. It uses a vision encoder trained from scratch.

  6. June 2025

    Magistral reasoning model

    In June 2025, he coauthors Magistral, Mistral's first reasoning model, and presents a scalable reinforcement-learning pipeline built on Mistral's models and infrastructure.

  7. July 2025

    Voxtral audio models

    In July 2025, he coauthors Voxtral Mini and Voxtral Small, multimodal audio chat models that understand speech and text. Their 32K context window supports audio files up to 40 minutes.

  8. As of October 2026

    Speech synthesis at Mistral

    He is a Research Scientist at Mistral AI, where he works on open-source multimodal large language models such as Pixtral and Voxtral. In March 2026, he coauthors Voxtral TTS, an expressive multilingual text-to-speech model that generates natural speech from as little as 3 seconds of reference audio.

Career

  1. NowMistral AIResearch Scientist
  2. 2025–2026Institute of Electrical and Electronics Engineers
  3. 2021–2024Jadavpur University
  4. 2019–2021Carnegie Mellon University
  5. Google DeepMindResearcher

Education

  1. 2018Carnegie Mellon UniversityM.S. in Computational Data Science (Analytics)
  2. 2016Nanyang Technological UniversityB.Eng. and M.Sc.

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Soham Ghosh. 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. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Soham Ghosh. 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.
  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. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staffIncludes Mistral AI. 5,000 frontier users would be 38.2% of the core labs' 13,100 reported staff. We test the estimate against 3,913 named model contributors and dated all-staff records across Google DeepMind, OpenAI and Anthropic, plus six additional frontier-grade labs.

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 Soham Ghosh yet. Send us a free-licensed photo.

  1. Soham Ghosh sohamghosh121.github.io. Checked 20 October 2026.
  2. ExCL: Extractive Clip Localization Using Natural Language Descriptions aclanthology.org. Checked 20 October 2026.
  3. TIPS: Text-Image Pretraining with Spatial Awareness arxiv.org. Checked 20 October 2026.
  4. Pixtral 12B arxiv.org. Checked 20 October 2026.
  5. Magistral arxiv.org. Checked 20 October 2026.
  6. Voxtral arxiv.org. Checked 20 October 2026.
  7. Voxtral TTS arxiv.org. Checked 20 October 2026.