All news

Matching stories Newest first

1. Oct 20
   HHugging Face PapersResearch

   OpenVLHarness adds persistent multimodal memory and tool use to vision-language models

   The harness gives multimodal models a tool capability layer and persistent memory for images, data and web references. 6 point average gains over no-tool baselines across 22 benchmarks and eight backbones, including GPT-6 Luna at lower token cost.

   • Model: Qwen3-VL-8B
   • Model: Kimi K3
   • Model: GPT-6 Astra
   • Model: GPT-6 Luna
   • Model: Codex
   • Model: Visual Sketchpad
   • Topic: research
   • Topic: agents
2. Oct 20
   HHugging Face PapersResearch

   Hugging Face paper introduces HarnessPro for cheaper, stronger AI agent harnesses

   HarnessPro evolves agent harnesses without ground-truth tests and outputs lite, medium and heavy variants. 1% versus RHO.

   • Model: GPT-5.5
   • Model: GLM-5.2
   • Model: Codex
   • Model: Claude Code
   • Topic: research
   • Topic: agents
   • Topic: coding
   • Topic: benchmark