RL Frontiers Engineer: Scale-Driven Research & Systems

Alex Loftus; New York (NY); Hybrid; USD 500,000 - 850,000; Full time

Benefits offered by this job

Equity donation matching

Flexible hours

Vacation and parental leave

Office in San Francisco

Job summary

Anthropic is seeking a Research Engineer / Research Scientist for RL Frontiers to develop next-generation RL architectures and algorithms, scaling from small-scale results to frontier-scale runs, and building the systems that enable fast, reliable experiments.

You’ll work across research and engineering on large language models, with Python and JAX or PyTorch, and contribute to reproducible experiments, performance modeling, and scalable infrastructure for throughput and stability at scale.

Qualifications

• Bachelor’s degree or equivalent education or experience.
• Strong background in ML/RL research and large-scale training.
• Experience with Python, JAX or PyTorch in a research context.

Responsibilities

• Study RL training and scaling with model size, context, and compute.
• Develop next-generation RL architectures and algorithms at frontier scale.
• Diagnose scaling differences due to numerical, algorithmic, or systemic causes.
• Build experimental infrastructure for fast, reproducible large-scale comparisons.
• Own end-to-end performance of largest RL runs, from code to hardware.
• Develop performance and cost models to guide architectural choices.
• Investigate training dynamics, instabilities, and throughput regressions.
• Maintain deep familiarity with transformer language models and scaling behavior.
• Gain hands-on experience training large models in distributed settings.

Skills

Python

JAX

PyTorch

RL research experience

Distributed training

Education

Bachelor's degree

Tools

Python

JAX

PyTorch

Job description

Anthropic is seeking a Research Engineer / Research Scientist for RL Frontiers to develop next-generation RL architectures and algorithms, scaling from small-scale results to frontier-scale runs, and building the systems that enable fast, reliable experiments.

You’ll work across research and engineering on large language models, with Python and JAX or PyTorch, and contribute to reproducible experiments, performance modeling, and scalable infrastructure for throughput and stability at scale.