PeopleofSuperintelligence

Researcher

Kush Jain

Kush Jain is an AI research scientist at Mistral AI, where he works on the Devstral models and coding agents. He studied at UT Austin, where he built the mCoq mutation analysis tool for Coq with Milos Gligoric, and then earned a software engineering PhD at Carnegie Mellon under Claire Le Goues. His doctoral research brought language models to software testing, with the CAT-LM model in 2023, the TestGenEval benchmark in 2024 and a 2025 LLM4Code best paper award for work on memorized bug benchmarks. At Mistral, he coauthored Magistral, the company's first reasoning model, in June 2025, followed by Voxtral, Devstral-Small and Ministral 3. In April 2026, he gave a keynote at LLM4Code arguing that academic research still matters for building coding agents.

Works
32
Citations
128
h-index
5
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2020–2021

    Mutation testing at UT Austin

    At UT Austin, he works with Milos Gligoric on mutation testing and test generation, applying mutation testing to formal verification to detect incomplete specifications. His tool paper mCoq: Mutation Analysis for Coq Verification Projects appears at ICSE 2020, and in 2021 he coauthors an OOPSLA paper on parallel bounded exhaustive testing.

  2. 2022–2023

    PhD research at Carnegie Mellon

    He becomes a software engineering PhD student in the Software and Societal Systems Department at Carnegie Mellon University, advised by Claire Le Goues, working where test generation meets large language models. In 2022, he coauthors Looking for Lacunae in Bitcoin Core's Fuzzing Efforts at ICSE 2022. MutationBERT, his predictive mutation testing approach that encodes both the mutated method and the test method, appears at ESEC/FSE 2023.

  3. 2023

    The CAT-LM model

    In 2023, he, Nikitha Rao, Uri Alon, Claire Le Goues and Vincent Hellendoorn present CAT-LM, a GPT-style language model with 2.7 billion parameters trained on Python and Java projects. Its pretraining signal accounts for the mapping between code files and their test files, and the paper appears at ASE 2023.

  4. 2024–2025

    The TestGenEval benchmark

    In October 2024, he, Gabriel Synnaeve and Baptiste Rozière release TestGenEval, a benchmark of 68,647 tests from 1,210 code and test file pairs across 11 Python repositories. GPT-4o, the strongest model tested, reaches an average coverage of 35.2%, and the paper appears at ICLR 2025.

  5. 2025

    TestForge and a best paper award

    In March 2025, he and Claire Le Goues present TestForge, an agentic framework that refines LLM-generated unit tests using feedback from test executions and coverage reports; on TestGenEval it reaches a pass@1 rate of 84.3%. A paper he coauthors on whether large language models memorize bug benchmarks wins the Best Paper Award at the LLM4Code workshop, held with ICSE 2025 in Ottawa in May 2025.

  6. June–July 2025

    Magistral and Voxtral at Mistral

    In June 2025, he is among the authors of Magistral, Mistral's first reasoning model, whose Medium version is trained with reinforcement learning alone on top of Mistral Medium 3. Voxtral Mini and Voxtral Small, two multimodal audio chat models released under the Apache 2.0 license, follow in July 2025 with him as a coauthor.

  7. August 2025–January 2026

    Devstral and Ministral 3

    In August 2025, he coauthors the technical report on Devstral-Small, a 24B open-source model for code agents. Ministral 3, which he coauthors in January 2026, is a family of parameter-efficient dense language models in 3B, 8B and 14B sizes.

  8. As of October 2026

    Research scientist at Mistral AI

    After completing his software engineering PhD at Carnegie Mellon, he is an AI research scientist at Mistral AI, working on Devstral and coding agents through synthetic data generation, human data annotation and reinforcement learning environments. In April 2026, he gives the LLM4Code keynote Beyond the Scaling Race: The Problems Academia Was Built to Solve at ICSE 2026 in Rio de Janeiro, arguing from Mistral's work on Devstral and Mistral Vibe that academic AI research stays relevant.

Career

  1. Since 2025Mistral AIAI research scientist
  2. 2020–2021University of Texas at Austin

Education

  1. 2023Carnegie Mellon UniversityPhD

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

Most-cited works

  1. CAT-LM Training Language Models on Aligned Code And Tests2023 · IEEE/ACM International Conference on Automated Software Engineering (ASE) · 46 citations
  2. Are Large Language Models Memorizing Bug Benchmarks?2025 · 15 citations
  3. Contextual Predictive Mutation Testing2023 · ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE) · 11 citations
  4. Mind the Gap: The Difference Between Coverage and Mutation Score Can Guide Testing Efforts2023 · IEEE International Symposium on Software Reliability Engineering (ISSRE) · 11 citations
  5. Syntax Is All You Need: A Universal-Language Approach to Mutant Generation2024 · Proceedings of the ACM on software engineering. · 10 citations

From OpenAlex, as of 21 October 2026.

  1. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Kush Jain. 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 Kush Jain. 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 Kush Jain yet. Send us a free-licensed photo.

  1. About me - Kush Jain kushjain.com. Checked 20 October 2026.
  2. Publications - Kush Jain kushjain.com. Checked 20 October 2026.
  3. Beyond the Scaling Race: The Problems Academia Was Built to Solve (LLM4Code 2026) - ICSE 2026 conf.researchr.org. Checked 20 October 2026.
  4. [2309.02389] Contextual Predictive Mutation Testing arxiv.org. Checked 20 October 2026.
  5. [2310.01602] CAT-LM: Training Language Models on Aligned Code And Tests arxiv.org. Checked 20 October 2026.
  6. [2410.00752] TestGenEval: A Real World Unit Test Generation and Test Completion Benchmark arxiv.org. Checked 20 October 2026.
  7. [2503.14713] TestForge: Feedback-Driven, Agentic Test Suite Generation arxiv.org. Checked 20 October 2026.
  8. CMU Portugal students win Best Paper Award at international workshop - CMU Portugal cmuportugal.org. Checked 20 October 2026.
  9. [2506.10910] Magistral arxiv.org. Checked 20 October 2026.
  10. [2507.13264] Voxtral arxiv.org. Checked 20 October 2026.
  11. [2509.25193] Devstral: Fine-tuning Language Models for Coding Agent Applications arxiv.org. Checked 20 October 2026.
  12. [2601.08584] Ministral 3 arxiv.org. Checked 20 October 2026.