Benchmark Almanac

GLM-4.1V-9B-Thinking

GLM-4.1V-9B-Thinking places in the top three on CrossVid.

Z.ai Released July 1st, 2025 Family GLM-V The lab's model card Compare with others
48
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
Z.ai Jul 2025

GLM-4.1V-9B-Thinking

Open weights Small

129.2 #281 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • MMMU-Pro 57.1% #70 /90
  • ERQA 45.8% #32 /36
  • MMVU 59.4% #23 /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
HallusionBench 63.2% #19 of 31 Lab model card
MMBench-CN v1.1 84.7% #4 of 5 Lab model card

Coding

BenchmarkScoreRankSource
SArena 0.82 #12 of 17 Official leaderboard
Flame-React-Eval 72.5% #4 of 4 Lab model card

Agents and tools

BenchmarkScoreRankSource
OSWorld 14.9% #12 of 13 Lab model card
SnakeBench 13 #179 of 188 Official leaderboard
AndroidWorld 41.7% #22 of 23 Lab model card
GameHorizon 55.0 #36 of 37 Official leaderboard
WebVoyager 69% #12 of 12 Lab model card

Vision, audio and robotics

BenchmarkScoreRankSource
CrossVid 35.1% #2 of 13 Official leaderboard
MMMU 68% #41 of 136 Official leaderboard
MM-HELIX 23.7% #6 of 17 Official leaderboard
MathVista 80.7% #28 of 70 Lab model card
CVBench 85% #6 of 15 Lab model card
BLINK 65.1% #14 of 29 Lab model card
Math-VR 29% #14 of 26 Official leaderboard
AI2D 87.9% #30 of 52 Lab model card
MuirBench 74.7% #7 of 12 Lab model card
CompassArena Multimodal 1,073 #5 of 8 Official leaderboard
MMStar 72.9% #23 of 36 Lab model card
MME-CC 19.3% #8 of 12 Official leaderboard
ZeroBench 19.2% #21 of 31 Lab model card
MMBench-EN v1.1 85.8% #16 of 23 Lab model card
MMLongBench-Doc 42.4% #12 of 17 Official leaderboard
LogicVista 60.4% #10 of 14 Lab model card
MathVerse 68.4% #9 of 12 Lab model card
RefCOCO / RefCOCO+ / RefCOCOg 87.4% #22 of 29 Lab model card
Video-MME 73.6% #39 of 51 Lab model card
MMMU-Pro 57.1% #70 of 90 Lab model card
OCRBench 84.2% #28 of 34 Lab model card
MATH-Vision (MathVision) 54.4% #43 of 51 Lab model card
We-Math 63.8% #11 of 13 Lab model card
MVBench 68.4% #20 of 23 Lab model card
MMVU 59.4% #23 of 26 Lab model card
ERQA 45.8% #32 of 36 Lab model card
VideoMME v2 (with subtitles) 14% #29 of 31 Official leaderboard
Video-MMMU 61% #34 of 36 Lab model card
DynaMath 42.5% #22 of 23 Lab model card
LVBench 45.1% #30 of 31 Lab model card
MotionBench 59% #9 of 9 Lab model card
ChartQA Pro 59.5% #8 of 8 Lab model card
TreeBench 37.5% #7 of 7 Lab model card
All-Angles Bench 52.7% #6 of 6 Lab model card
OmniSpatial 47.7% #6 of 6 Lab model card
ChartMuseum 48.8% #5 of 5 Lab model card
Ref-L4 86.8% #3 of 3 Lab model card
WebQuest 63.4% #3 of 3 Lab model card
GeoBench 76% #2 of 2 Lab model card

Its lab's numbers put it at 129.2, number 281 of 404

The Self-Reported Index fits one score to the 28 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 127.3 to 131.2. How we compute it.

Its lab's numbers next to 5 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
MotionBench 59% MotionBench 59.4% Leaderboard, frames: 2fps −0.4 pts
MATH-Vision (MathVision) 54.4% MathVision, GLM-4.1V-9B-Thinking thinking (Table 2) 54.4% Leaderboard Same
MMLongBench-Doc 42.4% MMLongBench-Doc 42.4% Leaderboard, VL; 9B Same
MMMU 68% MMMU 68% Leaderboard, thinking Same
MMMU-Pro 57.1% MMMU-Pro 57.1% Leaderboard, thinking Same

Its launch post

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

GLM-4.1V-9B-Thinking model card and technical report Jul 1, 2025

Its system card

Every result in the longer documents the lab published with it. 14 of them aren't in the launch post.

GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning Jul 1, 2025

Capabilities