UCLA x Apple open-source OpenVLHarness multimodal agent, gaining 9.7-12.6 points across 22 benchmarks
WeijiaShi2 · x · 2026-10-22
The UCLA × Apple team introduces OpenVLHarness, a multimodal agent that inspects, computes over, and reasons about evidence, featured by AK.
Technical core:
- A capability layer gives consistent interfaces for tools like perception and search, rendering outputs into model-inspectable observations
- Persistent multimodal memory retains images, masks, numerical arrays, and web references under reusable handles
- A coding agent computes over these artifacts to combine evidence across tools
Across 22 benchmarks and 8 backbones, it gains 9.7–12.6 points on average over matched no-tool baselines, showing evidence acquisition and extra reasoning are complementary.
More from coding & agent
- OpenLore gives every coding agent the same memory with one folder of Markdown — Arindam_1729 · 2026-10-22
- Walking bento box built with Claude Code + Blender Python, no MCP needed — sidahuj · 2026-10-22
- Claude Code Autonomously Generated 2,000+ Krea2 Images Overnight — reynadsaltynuts · 2026-10-22
- Indie dev building game #4 with AI starts with a central game asset management hub — ezshine · 2026-10-22
- Diffy: a cross-platform desktop code review app with a custom semantic diff engine — seatedro · 2026-10-22
- Open-Source Garmin MCP Connector Plugs Workout Data Into AI Agents — Vast-Grapefruits · 2026-10-22