PeopleofSuperintelligence

Builder

Xuanyu Zhang

Xuanyu Zhang is a software engineer at Mistral AI whose work spans language model inference, reasoning, coding agents and multimodal audio. While at Apple in 2024, he coauthored Recurrent Drafter (ReDrafter) for speculative decoding and contributed to integrating it into NVIDIA TensorRT-LLM in December 2024. In 2025, he coauthored the Magistral reasoning models, the Voxtral multimodal audio chat models and the Devstral-Small model for coding agents. At a Munich vLLM meetup in February 2026, he presented on Mistral AI and vLLM, and in July 2026 he contributed a FlashInfer monolithic MoE routing replay change to vLLM.

Timeline

  1. 2024

    Recurrent Drafter at Apple

    While at Apple in 2024, he coauthors Recurrent Drafter (ReDrafter), a speculative decoding method for large language model inference. The approach pairs a recurrent neural network draft model conditioned on hidden states with dynamic tree attention over beam search results.

  2. December 2024

    ReDrafter in TensorRT-LLM

    In December 2024, he contributes to Apple's collaboration with NVIDIA to integrate ReDrafter speculative decoding into the NVIDIA TensorRT-LLM inference acceleration framework.

  3. June 2025

    Magistral reasoning models

    He coauthors the June 2025 Magistral paper on Mistral's first reasoning model, which presents Magistral Medium trained on top of Mistral Medium 3 with reinforcement learning alone and open-sources Magistral Small under Apache 2.0.

  4. July 2025

    Voxtral audio chat models

    The July 2025 Voxtral report, which he coauthors, presents Voxtral Mini and Voxtral Small as multimodal audio chat models trained to comprehend spoken audio and text documents.

  5. 2025

    Devstral for coding agents

    Zhang coauthors the 2025 Devstral paper, which introduces Devstral-Small, a lightweight open-source language model fine-tuned for coding agent applications.

  6. February 2026

    Software engineer at Mistral AI

    At a Munich vLLM meetup in February 2026, he is a Software Engineer at Mistral AI and presents a talk titled 'Mistral AI & vLLM' alongside Patryk Saffer.

  7. July 2026

    FlashInfer MoE routing in vLLM

    In July 2026, he contributes a merged vLLM change that adds routing_replay_out support to the FlashInfer monolithic MoE backend so RoutedExpertsCapturer can capture expert routing decisions.

Career

  1. From 2026Mistral AISoftware Engineer
  2. 2024Apple

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Xuanyu Zhang. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Mistral AI. 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.
  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 with dated positions at the same organization in the same years.

Sources and credits

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

  1. Recurrent Drafter for Fast Speculative Decoding in Large Language Models arxiv.org. Checked 22 October 2026.
  2. Accelerating LLM Inference on NVIDIA GPUs with ReDrafter machinelearning.apple.com. Checked 22 October 2026.
  3. Magistral arxiv.org. Checked 22 October 2026.
  4. Voxtral arxiv.org. Checked 22 October 2026.
  5. Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 22 October 2026.
  6. Munich vLLM Meetup · Luma luma.com. Checked 22 October 2026.
  7. Enable router replay output from FlashInfer monolithic MoE kernel · #44214 github.com. Checked 22 October 2026.