Skip to main content
The following is a list of recent changes to the VLM Run Platform and API.
​
2026-06-26
v2026-06-26
  • Orion 2 Release: Full launch of Orion 2, our most capable visual agent, now with code execution.
  • Model Selector: Choose which model runs inside the Orion 2 visual agent harness, from open-weight models like Gemma 4 to closed models like Opus 4.8.
  • Media Library: Upload, browse, and chat with your own media and documents through a new library panel in chat.
  • Healthcare multi-document workflow cookbook: Learn how to take multiple documents, classify them, and extract information from a specific document.
​
2026-04-27
v2026-04-27
  • New Platform tab: a single place to learn how the VLM Run Platform works.
  • Service Tiers (Standard/Flex/Priority): a new service_tier parameter is now supported end-to-end: from the backend, through the Python SDK’s GenerationConfig, to full documentation with cURL and SDK examples.
  • New evaluations guide: learn how to measure skill accuracy, collect feedback, run evaluations, and use results to optimize your skills over time.
  • Improved video processing and transcription with longer, richer output.
  • Process specific document pages with the new page_indices parameter in GenerationConfig.
  • LaTeX rendering in chat messages for math and scientific content.
  • Lossless PNG support for segmentation masks with color keys, unlocking more segmentation use cases.
​
2026-03-26
v2026-03-26
  • Evaluations: Measure workflow improvements using skills in the Evaluations tab.
  • Skills:
    • Support for uploading skills via zip files
    • Refer to skills quickly in chat by typing / and selecting a skill you’ve used
    • Python and Node SDKs support skills management, inline skills, and skill-driven extraction
  • Improved chat completion latency
  • Improved multimodal content handling
​
2025-04-25
v0.5.0
  • New /document/execute endpoint for running arbitrary VLM DAGs over documents.
  • Added support for document.markdown domain for extracting structured markdown from documents.
​
2025-03-19
v0.4.0
  • Fixes for multi-page grounding support with custom JSON schemas in Document -> JSON.
  • Added support for audio.transcription domain for long-form audio transcription (upto 12 hours of audio).
  • Added support for video.transcription domain for long-form video transcription (upto 3 hours of video).
  • Added cookbook for video transcription.