Jump Trading Scales Quant Research with GPT-6 Astra for Long-Horizon Autonomous Tasks

Jump Trading is using GPT-6 Astra to expand the scope of its quantitative research, moving beyond simple code snippets to comprehensive, long-running analytical tasks. Lucas Baker, Head of LLM R&D at Jump, notes that the model unlocks a new tier of autonomy for long-horizon tasks that require flexible agent coordination and persistence, reducing the need for frequent human guidance.

Recursive Improvement and Autonomy

The firm treats AI agents as colleagues capable of recursive improvement, where the system analyzes its own findings against initial criteria and redirects its efforts without human intervention for each step. This allows agents to handle complex workflows that span days, pulling from multiple data sources and interrelating findings to create comprehensive analyses.

Safety in Regulated Environments

Given the strict regulatory environment of finance, Jump Trading maintains human judgment at the center of its workflow. Agents operate within secure, well-monitored environments with clear constraints, and all outputs—such as trading signals—undergo human review and validation before integration into execution systems. This approach ensures that while agents can produce any output needed, critical validation prevents financial and compliance risks.

Future of Autoresearch

Baker envisions a future where autoresearch—the recursive improvement of measurable systems by agent researchers—becomes a standard part of a quant researcher's workflow. While current implementations still require regular check-ins, the goal is a loosely structured fleet of agents coordinated by other agents, capable of exploring open questions and allocating compute time with minimal human-defined inputs.