Ege Erdil
Ege Erdil is an AI researcher whose work connects model progress, mathematical reasoning, and the economics of automation. At Epoch AI, he studied algorithmic gains in computer vision and language models, then coauthored FrontierMath, a benchmark for advanced mathematical reasoning. He coauthored GATE, a model of AI automation’s economic effects, and left Epoch AI in April 2025 to launch Mechanize with colleagues. In May 2026, he coauthored GBA Eval, which asks models to build a Game Boy Advance emulator in Rust. He serves as Mechanize’s CTO in June 2026, discussing evaluations.
Timeline
-
2022
Algorithmic progress in vision
In 2022, he coauthors a study of more than 100 computer-vision models and estimates that better algorithms add the equivalent of doubled compute budgets every nine months.
-
March 2024
Language model efficiency
In March 2024, he coauthors a study that finds algorithmic progress in language models matches a doubling of computational power every five to 14 months.
-
November 2024
FrontierMath benchmark
In November 2024, he coauthors FrontierMath, a benchmark of original, difficult math problems created and vetted by mathematicians to evaluate advanced mathematical reasoning in AI.
-
March 2025
Modeling AI automation
In March 2025, he coauthors GATE, a dynamic integrated model that simulates the economic effects of AI automation across possible scenarios.
-
April 2025
Launching Mechanize
In April 2025, he leaves Epoch AI with colleagues to launch Mechanize, a startup developing virtual work environments, benchmarks, and training data that help AI agents learn from realistic job tasks.
-
May 2025
Automating software engineering
In May 2025, he coauthors a Mechanize essay about fully automating software engineering and the roles of human training data and reinforcement learning in deep learning.
-
May 2026–June 2026
Mechanize CTO and GBA Eval
In May 2026, he coauthors GBA Eval, which asks models to build a Game Boy Advance emulator in Rust. In June 2026, he serves as Mechanize’s CTO and discusses evaluation design.
Sources and credits
Photo: We have no published photo of Ege Erdil yet. Send us a free-licensed photo.
- Revisiting algorithmic progress
- Algorithmic progress in language models
- FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI
- GATE: An Integrated Assessment Model for AI Automation
- The Epoch AI Brief - April 2025
- Announcing Mechanize, Inc.
- How to fully automate software engineering
- Introducing GBA Eval
- How bad data teaches models to write terrible code
