Particle.news

OpenAI Upgrades Operator to o3 Model for Enhanced Autonomy and Safety

The advanced o3 reasoning model improves Operator’s task performance, safety measures, and clarity, remaining exclusive to $200/month ChatGPT Pro subscribers.

Overview

  • Operator, OpenAI's semi-autonomous web-browsing agent, now uses the o3 reasoning model, replacing the previous GPT-4o version.
  • The o3 upgrade enhances task success rates, accuracy, and structured outputs, with benchmarks showing significant performance improvements over its predecessor.
  • Enhanced safety measures reduce susceptibility to prompt injection attacks and improve compliance with safety protocols for sensitive actions.
  • Operator continues to operate in a cloud-hosted virtual browser, maintaining user privacy and security while completing browser-based tasks.
  • The upgraded Operator remains in research preview, accessible only to ChatGPT Pro subscribers, with API access still based on GPT-4o for now.