zai-org

/

GLM-4.6V

like 297

Follow

Z.ai 5.9k

Added structured evaluation results from benchmark comparison (36 benchmarks):

General VQA:

• MMBench V1.1: 88.8, MMBench V1.1 (CN): 88.2, MMStar: 75.9, BLINK (Val): 65.5, MUIRBENCH: 77.1

Multimodal Reasoning:

• MMMU (Val): 76.0, MMMU_Pro: 66.0, VideoMMU: 74.7, MathVista: 85.2, AI2D: 88.8, DynaMath: 54.5, WeMath: 69.8, ZeroBench (sub): 25.8

Multimodal Agentic:

• MMBrowseComp: 7.6, Design2Code: 88.6, Flame-React-Eval: 86.3, OSWorld: 37.2, AndroidWorld: 57.0, WebVoyager: 81.0, Webquest-SingleQA: 79.5, Webquest-MultiQA: 59.0

Multimodal Long Context:

• MMLongBench-Doc: 54.9, MMLongBench-128K: 64.1, LVBench: 59.5

OCR & Chart:

• OCRBench: 86.5, OCR-Bench_v2 (EN): 65.1, OCR-Bench_v2 (CN): 59.6, ChartQAPro: 65.5, ChartMuseum: 58.4, CharXiv_Val-Reasoning: 63.2

Spatial & Grounding:

• OmniSpatial: 52.0, RefCOCO-avg (val): 88.6, TreeBench: 51.4, Ref-L4-test: 88.9

This enables the model to appear in leaderboards and makes it easier to compare with other models.