Context Researcher for Frontier Agent Training

OpenAI; San Francisco (CA); On-site; USD 380,000 - 500,000; Full time

Job summary

OpenAI is seeking a Context Researcher for the Agent Post-Training team in San Francisco. You will scale compute spent on context, work across Codex Chronicle and frontier training stacks, and collaborate with researchers, engineers, product, and safety teams to influence major model runs and product outcomes.

You will design experiments, own end-to-end improvements to the post-training stack, and help build evaluative tools to detect and address model failures in production-like environments.

Qualifications

• Strong foundation in ML, software engineering, systems, or statistics.
• Hands-on experience with LLMs, RL, RLHF/RLAIF, and post-training workflows.
• Familiarity with evaluation tooling and production ML systems.

Responsibilities

• Design experiments to scale compute on context for agents.
• Own end-to-end improvements to the post-training stack including RL, data pipelines, and graders.
• Build evals and environments to uncover model failures and drive improvements.
• Collaborate with Codex, ChatGPT product teams to translate signals into model changes.
• Work on early-training and alignment interventions; shape downstream agent behavior.
• Improve training and launch machinery for velocity, observability and cost.

Skills

Machine learning

Software engineering

Systems design

Statistics

RL/RLHF/RLAIF

LLM tooling

Tools

Code pipelines

Eval frameworks

Data pipelines

Job description

OpenAI is seeking a Context Researcher for the Agent Post-Training team in San Francisco. You will scale compute spent on context, work across Codex Chronicle and frontier training stacks, and collaborate with researchers, engineers, product, and safety teams to influence major model runs and product outcomes.

You will design experiments, own end-to-end improvements to the post-training stack, and help build evaluative tools to detect and address model failures in production-like environments.