← Leaderboard

Qwen3-VL 30B-A3B Instruct 4bit MLX

Tested Rank #20 of 33 · 3/8 GAUNTLET progress
Generalist — pending G Agentic — pending A Understanding — 81.7 / 100 U Needle — pending N Thinking — 100 / 100 T Live — pending L Engineering — pending E Throughput — 30.2 / 100 T

Hollow markers & dashed spokes: axis not yet scored

G
A
U
81.7
N
T
100
L
E
T
30.2

Specification

Parameters
30B total / 3B active per token
Architecture
qwen3_vl_moe
Size on disk
18 GB
Quantization
4bit
Format
MLX
Reasoning (CoT)
No
Internal ID
M23
Mean speed
49.2 tok/s across suites
Stall census
0 stalls in 17 observed tests (0.0%)
Model card
lmstudio.ai

Suite results

Suite Score Avg / 20 Tests tok/s
Doc/OCR vision 1d 151.2 / 160 18.9 8/8 53.4
Doc/OCR — real-degraded tier 1d2 44.8 / 80 11.2 4/4 45.1

Each test is LLM-judged 0–20 against a fixed rubric; suite max = tests × 20. Where fewer tests ran than the suite total, unrun tests count as zero toward the axis score. See methodology.