Skip to main content
October 12, 2026 This release introduces an Agent Evaluations tab and gives livestream agents richer evidence to work with. It also adds support for new GPT-6 models and tightens review and policy reliability.

What’s new

  • Agents Livestream agents can now reverse-search their own frames and see the matches.
  • Agents Merchant risk now keeps marketplace reviews out of Public Reviews.
  • Platform A new Agent Evaluations tab covers datasets, evaluators, and experiments.
  • Platform Evaluators can now return a score, choices, or custom JSON.
  • Platform Evaluations can now score by section: policy, output, and trace parts.
  • Platform The PDF extraction tool is now billed at $0.02 per page.
  • Integrations Agents can now list a Google Drive folder’s subfolders.
  • Integrations Sardine connection setup now shows the public webhook URL.
  • QA Policy copy, description editing, and Markdown paste now work correctly.
  • QA Entering review mode no longer freezes the page.
  • Dashboard Support agent cards now show Recent conversations and About.
  • Models GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are now available.
  • API The jobs API now supports a review status filter and counts.
  • Workflows Skill set generation now shows why a run failed.