Sagar Vaze

Sagar Vaze is a researcher whose path runs from computer-vision representation learning at Oxford to multimodal language models at Mistral AI. At the University of Oxford, he did his PhD in the Visual Geometry Group under Andrew Zisserman and Andrea Vedaldi, published 2022 papers on open-set recognition (ICLR) and Generalized Category Discovery (CVPR), and spent time at Meta AI, including the Segment Anything team. He co-authored Pixtral 12B in 2024, Magistral, Mistral's first reasoning model, and Meta's SAM 3 in 2025, and Ministral 3 in 2026. He is a Research Scientist at Mistral AI, where he leads the Multimodal team.

Now at: Mistral AI
Works: 25
Citations: 689
h-index: 8
Badges: Bridge

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

By 2022

A PhD at Oxford's VGG

By 2022, he is a PhD student in the Visual Geometry Group (VGG) at the University of Oxford, working on representation learning in computer vision under Andrew Zisserman and Andrea Vedaldi. During the PhD he also spends time at Meta AI, first with Ishan Misra in New York and then in the Segment Anything team with Ross Girshick.
2022

Open-set recognition and category discovery

In 2022, he is first author of two papers with Kai Han, Andrea Vedaldi and Andrew Zisserman. Open-Set Recognition: a Good Closed-Set Classifier is All You Need? is an oral at ICLR 2022 and shows that a classifier's 'none-of-above' decision is highly correlated with its accuracy on the closed-set classes. Generalized Category Discovery appears at CVPR 2022 and introduces a semi-supervised k-means method that clusters unlabelled images into seen and unseen classes.
2023

Clevr-4 for category discovery

His NeurIPS 2023 paper No Representation Rules Them All in Category Discovery, written with Andrea Vedaldi and Andrew Zisserman, introduces Clevr-4, a synthetic dataset with four equally valid partitions of the data by object shape, texture, color or count. A mean-teacher method on the same task sets a new state of the art on the Semantic Shift Benchmark.
October 2024

Pixtral 12B at Mistral AI

In October 2024, he is among the 42 authors of Pixtral 12B, Mistral AI's 12-billion-parameter multimodal language model. Pixtral uses a new vision encoder trained from scratch, which ingests images at their natural resolution and aspect ratio.
June 2025

Mistral's first reasoning model

In June 2025, he is among the 99 authors of Magistral, Mistral AI's first reasoning model, which comes with its own scalable reinforcement learning pipeline. Magistral Medium is trained for reasoning on top of Mistral Medium 3 with reinforcement learning alone, and Magistral Small is open-sourced under Apache 2.0.
November 2025

SAM 3 with Meta's SAM team

In November 2025, he is among the 38 authors of SAM 3: Segment Anything with Concepts, a model that detects, segments and tracks objects in images and videos from concept prompts. The paper, credited to the SAM Team at Meta AI, appears at ICLR 2026.
January 2026

Ministral 3 model family

In January 2026, he is an author of Ministral 3, Mistral AI's family of dense language models in 3B, 8B and 14B sizes. Each size comes as a base, an instruction-tuned and a reasoning model, all with image understanding and an Apache 2.0 license.
As of October 2026

Multimodal team at Mistral

He is a Research Scientist at Mistral AI, where he leads the Multimodal team across pre-training and post-training, on topics such as science of data, robotics and multimodal agents. In July 2026, he is an author of Robostral Navigate, an 8B vision-language model that predicts navigation waypoints from monocular RGB images.