Thomas Scialom
Corresponding author of Meta's Llama 2 paper; a lead author of the ARE agent platform and Gaia2 benchmark
- Listed with GenAI, Meta on 2023 papers and with Meta Superintelligence Labs on a 2025 paper and a February 2026 paper (Gaia2), Meta1534
Thomas Scialom is one of the two equal-contribution corresponding authors of Llama 2: Open Foundation and Fine-Tuned Chat Models (2023), from GenAI at Meta, and is listed in that paper's Science and Engineering Leadership group; Scialom is also among the core contributors of The Llama 3 Herd of Models (2024). With Meta Superintelligence Labs, Scialom is one of the lead authors of ARE: Scaling Up Agent Environments and Evaluations (2025), which introduces the Meta Agents Research Environments platform and the Gaia2 benchmark. A 2026 paper, the Gaia2 benchmark paper published at ICLR 2026, lists Scialom at Meta SuperIntelligence Labs. Scialom also co-authored the earlier GAIA benchmark for general AI assistants (2023).12345
Work
Llama 2 and Llama 3Open weights
Equal-contribution corresponding author of Llama 2 (2023), listed under Science and Engineering Leadership in its contributions section, and a core contributor of The Llama 3 Herd of Models (2024), which says pre-trained and post-trained versions of the 405B model were publicly released.12
ARE and Gaia2 (external site: arxiv.org)Published research
One of four lead authors of ARE: Scaling Up Agent Environments and Evaluations (2025), which presents a research platform for building agent environments and the Gaia2 benchmark, and a co-author of Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments (ICLR 2026).34
GAIA (external site: arxiv.org)Published research
Co-author of GAIA: a benchmark for General AI Assistants (2023), with Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf and Yann LeCun.5
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.