Bio
Lucas Atkins serves as CTO and Head of Research at Arcee, where he led the development of the company’s proprietary training stack. He has worked in machine learning since 2019, including text-to-speech systems for automotive systems. His work also includes training specialized language translation systems for the UAE in 2022 and collaborating with ROCM from 2023 to 2024 on optimizing enterprise GPUs for large-scale model training.
Sessions (2)
Additional Speaker Details
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Lucas Atkins is the CTO (and head of research) at Arcee AI, where he leads the pretraining and architecture work behind the company's Trinity model family — including Trinity Large, a 400B-parameter (13B active) open-weight Mixture-of-Experts model trained on 17T tokens and released under Apache 2.0. He is the final author on the Trinity Large technical report, which introduces the SMEBU load-balancing method and reports training all three Trinity models with zero loss spikes, and he wrote Arcee's Trinity Manifesto. His talks and writing dig into the engineering of large-scale training stability, and he holds a public, data-grounded thesis that open models will keep closing the gap — predicting 'by the end of 2026 the best model in the world is some degree of open.'
Speaking style
Recent talks (2)
Recent writing (3)
- Trinity-Large-Thinking: Scaling an Open Source Frontier Agent · blog · Apr 2026
- Arcee Trinity Large Technical Report · paper · Feb 2026
- Announcing Arcee Foundation Models · blog · Jun 2025
Podcasts & interviews (1)
- The OpenRouter Show ft Lucas Atkins - Episode 4 (Arcee AI) · The OpenRouter Show · Feb 2026