ZwZ2 trains vision models to connect evidence across image regions

EvidenceConnect shows that verified single-image, multi-region questions used for RL training can improve spatial reasoning, long-context and GUI-agent benchmarks without multi-image data. The code, data and 4B/9B/27B models are released, so the recipe can be reproduced.

• context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
• AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

• postAnt Open Source ships quantized Ling-3.0-flash-Fin for finance workloads
• postAlipay-PIBench asks whether agents can integrate payments safely
• postLing-3.0-tiny opens weights in BF16, FP8 and INT4
• postPost-training a small Ling model with agentic RL on one DGX Spark

• articleZooming Without Zooming: Distilling Crop Tools into One Forward PassAnt Open Source
• articleInside GLM-5V-Turbo: dual-teacher vision distillation and 30-task RLZ.ai
• articleMiMo-VL-7B reports GUI grounding ahead of specialist modelsXiaomi MiMo

Checking sign-in…

Loading comments…