Jump Trading scales quantitative research using GPT-6 Astra
Jump Trading is utilizing GPT-6 Astra to automate complex, long-horizon quantitative research and coding workflows, enabling a shift from manual code generation to autonomous agentic research. This allows quantitative researchers to focus on defining high-level goals and evaluation criteria while agents handle the execution and recursive improvement of hypotheses.
Autonomous Agentic Workflows for Quantitative Research
GPT-6 Astra enables Jump Trading to execute long-horizon tasks that require flexible coordination between agents and high persistence. According to Lucas Baker, Head of LLM R&D at Jump, the model has unlocked a new tier of autonomy, reducing the need for frequent human intervention. Researchers now define the problem, the work environment, and the evaluation metrics, then steer agents in real time.
Key capabilities introduced by GPT-6 Astra include:
- Recursive Improvement: Agents can find practical changes, merge and stack those wins, and actively redirect their efforts based on agreed-upon criteria without requiring a human to analyze every round of changes.
- Comprehensive Analysis: The system can run for several days, pulling from multiple data sources and making qualitative judgments on the importance of data to create a comprehensive analysis.
- End-to-End Codebase Development: The workflow has evolved from writing one-off code snippets in 2024 to creating entire codebases from scratch in 2025, and now collaborating dynamically to progress on open research questions in 2026.
Risk Management and Human Oversight in Regulated Finance
To manage the financial and compliance risks inherent in quantitative trading, Jump Trading integrates agentic intelligence into quality, security, and monitoring systems rather than just feature addition. The firm employs a strategy of "safe environments" combined with stringent human review.
"If you have a safe environment where the agent or system is free to produce any output that it needs, but there is also a human review process at the end of it where critical validation takes place with human acceptance, that’s what gives us confidence,"
Lucas Baker explains that any output, such as a trading signal produced by an agent, is treated as a signal that is informative but potentially incorrect. These signals are integrated into a stringently reviewed and controlled execution environment to ensure safety and compliance.
The Future of "Autoresearch"
Jump Trading envisions a transition toward "autoresearch," where the recursive improvement of measurable systems by agent researchers becomes a standard part of the quantitative workflow.
In this future state, a human would define the initial system—including inputs, environment, evaluation metrics, and priorities—and then entrust the process to a "fleet" of agents coordinated by other agents. These agents would autonomously decide how to explore ideas, allocate compute resources, and integrate findings to answer general research questions. This shift aims to move the process from requiring regular check-ins to a fully end-to-end autonomous research pipeline.