MMM-Bench - Level 4 (Business Form): leaderboard

MMM-Bench (2026, Alibaba) zero-shot document classification at taxonomy level 4, Business Form, for real enterprise documents classified using the paper’s hierarchical category prompt. Macro-F1 in percent; higher is better.

Metric: Macro-F1 (%) at taxonomy level 4 (Business Form), zero-shot hierarchical classification of the test split (20 percent of 5,990 real Alibaba business documents in 12 domains, mostly Chinese) with a document prompt containing OCR text, image-page placeholders and optional filename; actual image-tensor input is described only for multimodal implementations against a flattened five-level taxonomy, temperature 0.1; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3.1 Pro (Preview)80.14#102
2Claude Opus 4.577.13#165
3GPT-5.475.1#138
4Claude Opus 4.671.4#114
5GPT-5.268.16#190
6Qwen 3.6 Plus63.96#245
7Qwen 3.5 27B57.74#380
8Qwen 3 VL 235B A22B57.53#486
9GLM-556.23#276
10Kimi K2.553.27#272
11Qwen 3.5 35B A3B51.89#494
12MiniMax-M2.748.62#539
13Qwen 3.5 9B38.82#838
14Qwen 3 VL 8B14.51#927
15Qwen 2.5 VL 7B0#1878

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=mmm-bench-level-4-business-form · How It Works · Data refreshed daily, snapshot 2026-11-22.