Benchmark Almanac

GLM-4.6V

GLM-4.6V places in the top three on 4 benchmarks.

Z.ai Released December 8th, 2025 Family GLM-V The lab's model card Compare with others
59
Benchmarks with a score
0
Ranked first, out of five or more models
4
In the top three
Z.ai Dec 2025

GLM-4.6V

Open weights Mid-size

136.8 #208 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • MMMU-Pro 66% #59 /90
  • ERQA 47.8% #30 /36
  • MMVU 68.4% #17 /26

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
LMArena (Elo) 1,378 #156 of 335 Official leaderboard
MMBench-CN v1.1 88.2% #3 of 5 Lab model card
HallusionBench 62.3% #20 of 31 Lab model card
Visual Aesthetic Benchmark 17.5% #17 of 18 Official leaderboard
MMLongBench 64.1% #4 of 4 Lab model card

Coding

BenchmarkScoreRankSource
Flame-React-Eval 86.3% #2 of 4 Lab model card

Agents and tools

BenchmarkScoreRankSource
SnakeBench 26 #66 of 188 Official leaderboard
OSWorld 37.2% #8 of 13 Lab model card
WebVoyager 81% #8 of 12 Lab model card
AndroidWorld 57% #21 of 23 Lab model card
GameWorld 30.8% #22 of 22 Official leaderboard
MM-BrowseComp 7.60% #5 of 5 Lab model card

Professional work

BenchmarkScoreRankSource
CADGenBench 18.3% #10 of 11 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
CVBench 87.6% #2 of 15 Lab model card
GSM8K-V 37.8% #5 of 24 Official leaderboard
MathVista 85.2% #16 of 70 Lab model card
OCRBench v2 65.1% #6 of 22 Lab model card
OmniSpatial 52% #2 of 6 Lab model card
WebQuest 69.3% #1 of 3 Lab model card
MMMU 76% #38 of 103 Lab model card
MVBench 74.9% #9 of 23 Lab model card
BLINK 65.5% #12 of 29 Lab model card
MuirBench 77.1% #5 of 12 Lab model card
WorldVQA 19% #10 of 22 Official leaderboard
AI2D 88.8% #24 of 52 Lab model card
MMLongBench-Doc 54.9% #8 of 17 Official leaderboard
MMStar 75.9% #17 of 36 Lab model card
Video-MME 81.8% #25 of 51 Lab model card
MADQA 66.1% #9 of 18 Official leaderboard
MathVerse 75.4% #6 of 12 Lab model card
ZeroBench 25.8% #16 of 31 Lab model card
MMBench-EN v1.1 88.8% #12 of 23 Lab model card
OCRBench 86.5% #19 of 34 Lab model card
TreeBench 51.4% #4 of 7 Lab model card
GeoBench 71% #28 of 48 Official leaderboard
RefCOCO / RefCOCO+ / RefCOCOg 88.6% #17 of 29 Lab model card
ChartMuseum 58.4% #3 of 5 Lab model card
ChartQA Pro 65.5% #5 of 8 Lab model card
LogicVista 62.1% #9 of 14 Lab model card
MMVU 68.4% #17 of 26 Lab model card
MMMU-Pro 66% #59 of 90 Lab model card
Ref-L4 88.9% #2 of 3 Lab model card
We-Math 69.8% #9 of 13 Lab model card
Vision Arena 1,161 #95 of 136 Official leaderboard
MATH-Vision (MathVision) 63.5% #37 of 51 Lab model card
DynaMath 54.5% #17 of 23 Lab model card
Video-MMMU 74.7% #27 of 36 Lab model card
MotionBench 63.6% #7 of 9 Lab model card
LVBench 59.5% #25 of 31 Lab model card
CharXiv 63.2% #61 of 75 Lab model card
BabyVision 17.6% #22 of 27 Official leaderboard
ERQA 47.8% #30 of 36 Lab model card
All-Angles Bench 56.2% #5 of 6 Lab model card
WorldBench 42.5% #7 of 8 Official leaderboard
SpatialWorld Digital 8.10% #8 of 9 Official leaderboard
PerceptionBench (Moonshot) 32.5% #16 of 16 Official leaderboard
DISBench 8.20% #9 of 9 Official leaderboard
SpatialWorld Physical 2.70% #9 of 9 Official leaderboard

Safety and security

BenchmarkScoreRankSource
nocot-bench (NCRI) 63.4 #127 of 185 Official leaderboard

Its lab's numbers put it at 136.8, number 208 of 404

The Self-Reported Index fits one score to the 30 capability results Z.ai reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 135.3 to 138.4. How we compute it.

Its lab's numbers next to 2 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
MATH-Vision (MathVision) 63.5% MathVision, GLM-4.6V thinking (Table 2) 63.5% Leaderboard Same
MMLongBench-Doc 54.9% MMLongBench-Doc 54.9% Leaderboard, VL; 12B activated (106B total) Same

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

GLM-4.6V blog and model card Dec 8, 2025

Its system card

Every result in the longer documents the lab published with it. 11 of them aren't in the launch post.

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning Jan 1, 2026

Capabilities