Anthropic
Research Engineer, Pretraining Scaling
$350K–$850K
San Francisco, CA +1 · Full-time · Posted Sep 30, 2025
- Location
- San Francisco, CA
San Francisco, California, United States - Experience
- Years required correlate with internal job level requirements; no specific number stated.
- Education
- Bachelor’s degree or an equivalent combination of education, training, and/or experience; relevant field of study demonstrated through coursework, training, or professional experience.
- Visa sponsorship
- Visa sponsorship is available, but not for every role or candidate; Anthropic says it will make reasonable efforts to obtain a visa after an offer.
About the job
Operate and improve Anthropic’s production pretraining systems so frontier models train reliably, efficiently, and at scale.
What you'll do
- Own and optimize production pretraining pipeline operations, performance, observability, and reliability.
- Debug training-stack issues and design experiments to improve efficiency, uptime, and model performance.
- Coordinate launch incident response and maintain training logs, monitoring, evaluation infrastructure, and codebase capabilities.
What they ask for
- Hands-on experience training large language models, or deep expertise with JAX, TPU, PyTorch, or large-scale distributed systems.
- Strong ability to debug complex, ambiguous problems across multiple stack layers.
- Bachelor’s degree or an equivalent combination of education, training, and/or experience; relevant field of study demonstrated through coursework, training, or professional experience.
Nice to have
- Contributed to open-source LLM frameworks such as open_lm, llm-foundry, or mesh-transformer-jax.
- Published research on model training, scaling laws, or ML systems.
Pay
$350,000 — $850,000 USD (base salary)
Tools
JAX, TPU, PyTorch, open_lm, llm-foundry, mesh-transformer-jax
AI focus
Train and scale production frontier models, with a focus on reliable, efficient pretraining and responsible scaling.
Model-written summary · the original posting is authoritative