# Eric Steinberger

> Eric Steinberger is the Austrian co-founder and CEO of Magic, a San Francisco AI company pursuing software engineering agents. A teenage reinforcement-learning paper led to collaboration with Noam Brown and research at Facebook AI; he also co-founded ClimateScience before starting Magic with Sebastian De Ro in 2022.

- **Role:** Co-founder and CEO at Magic
- **Organizations:** Magic, ClimateScience, Facebook AI Research
- **From:** Vienna, Austria
- **Nationality:** Austrian
- **Education:** Computer science studies; left after one year without a degree, University of Cambridge, Trinity College, Secondary education, HTL Spengergasse
- **Known for:** Authored Single Deep Counterfactual Regret Minimization and co-authored the DREAM reinforcement-learning paper with Noam Brown., Co-founded ClimateScience, a volunteer-powered climate education nonprofit., Co-founded Magic and led its work on long-context models for software engineering.

## Career timeline

- **2019** — Published Single Deep Counterfactual Regret Minimization, work that led to a collaboration with Noam Brown.
- **2019** — Began research at Facebook AI Research and co-founded ClimateScience.
- **2022** — Co-founded Magic with Sebastian De Ro.
- **2023** — Announced Magic's $23 million Series A and introduced LTM-1 with a 5 million token context window.
- **2024** — Magic announced LTM-2-mini with a 100 million token context window and a Google Cloud compute partnership.
- **2026** — Magic published research claiming more than tenfold pretraining compute efficiency versus leading open-weight base models.

## Achievements

- Authored Single Deep Counterfactual Regret Minimization and co-authored the DREAM reinforcement-learning paper with Noam Brown.
- Co-founded ClimateScience, a volunteer-powered climate education nonprofit.
- Co-founded Magic and led its work on long-context models for software engineering.
- Magic reported a 100 million token context model in 2024 and $515 million in total funding at that announcement.

## Latest updates

- **2026-09** — Magic reported a pretraining recipe it says is more than ten times as compute efficient as leading open-weight base models and said it would scale long-horizon reinforcement learning.
- **2024-08** — Magic announced LTM-2-mini, a 100 million token context model, Google Cloud infrastructure plans, and $515 million total funding.

## Links

- Website: https://magic.dev
- LinkedIn: https://www.linkedin.com/in/ericsteinb
- Twitter/X: https://x.com/EricSteinb
- GitHub: https://github.com/EricSteinberger

---

Profile page: https://yespress.io/eric-steinberger
Published by YesPress — https://yespress.io
Last updated: 2026-09-29