ICLR 2026: Scaling RL on Frontier LLMs with 1T+ Parameters
Author: @agarwl_ (Rishabh Agarwal)
Date: 2026-04-28
Summary
Talk at ICLR 2026 about scaling RL on frontier LLMs with 1T+ parameters, including experimental data from the field.
Key Insight
RL at 1T+ parameter scale — practical lessons from frontier model training.
Concepts
See Also
- making-rl-fast — RL speedups
- fireworks-frontier-rl-cheaper — cost reduction in RL