benchgap
Vision & documents

OfficeQA Pro leaderboard

As of 2026-11-21, the highest measured score on OfficeQA Pro is 72.9% by Claude Fable 5.2. 78 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.272.9%measured
2Claude Fable 5.169.0%measured
3Claude Opus 5.567.7%measured
4GPT-6.2 Astra67.6%estimated ± 6.8 pp, low confidence
5GPT-6 Astra67.6%estimated ± 6.8 pp, medium confidence
6GPT-6.1 Sol67.6%estimated ± 6.8 pp, medium confidence
7Gemini 3.8 Flash67.6%estimated ± 6.8 pp, medium confidence
8Gemini 3.7 Flash67.6%estimated ± 6.8 pp, medium confidence
9GPT-6.1 Terra67.6%estimated ± 6.8 pp, medium confidence
10Gemini 3.5 Flash67.6%estimated ± 6.8 pp, medium confidence
11GPT-5.6 Sol67.6%estimated ± 6.8 pp, medium confidence
12Gemini 3.6 Flash67.6%estimated ± 6.8 pp, medium confidence
13GPT-6 Sol67.5%estimated ± 6.8 pp, medium confidence
14Qwen3.8 Max Preview67.5%estimated ± 6.8 pp, medium confidence
15Kimi K3.1 Pro67.4%measured
16Gemini 3.1 Pro67.3%estimated ± 6.8 pp, medium confidence
17Claude Mythos 567.0%estimated ± 2.5 pp, medium confidence
18Claude Opus 566.9%measured
19Claude Opus 4.866.2%measured
20Hy4 preview66.2%measured
21Claude Sonnet 565.8%estimated ± 2.5 pp, low confidence
22DeepSeek V4.1 Flash 101765.7%estimated ± 2.9 pp, low confidence
23Claude Sonnet 5.565.6%measured
24Inkling 2 Small64.4%estimated ± 2.9 pp, low confidence
25Muse Spark Open63.4%estimated ± 6.8 pp, medium confidence
26Kimi K363.3%measured
27Hy463.1%estimated ± 7.0 pp, low confidence
28GLM-5.3-Flash62.4%measured
29GPT-5.6 Terra62.0%estimated ± 6.8 pp, medium confidence
30Inkling 2 Mini60.5%estimated ± 2.9 pp, low confidence
31Muse Spark60.4%estimated ± 6.8 pp, medium confidence
32Qwen3.7 Plus60.4%estimated ± 6.8 pp, medium confidence
33Step 5 Preview60.3%measured
34Grok 4.559.5%estimated ± 6.8 pp, medium confidence
35Claude Fable 557.9%measured
36Gemini 3 Pro57.8%estimated ± 6.8 pp, medium confidence
37GPT-6 Luna54.1%estimated ± 6.8 pp, medium confidence
38GPT-5.554.1%measured
39GPT-5.453.2%measured
40Kimi K2.652.6%estimated ± 6.8 pp, medium confidence
41Gemini 3.1 Flash-Lite52.3%estimated ± 7.0 pp, low confidence
42Apodex 1.151.9%estimated ± 6.8 pp, medium confidence
43Apodex 1.1 Mini51.9%estimated ± 6.8 pp, medium confidence
44Claude Haiku 5.551.9%estimated ± 7.0 pp, low confidence
45MiniMax M3.1 Flash51.7%estimated ± 6.8 pp, medium confidence
46Gemini 3.5 Flash-Lite51.4%estimated ± 6.8 pp, medium confidence
47Ling 3.0 Flash VL51.4%estimated ± 6.8 pp, medium confidence
48Gemini 3 Flash50.8%estimated ± 6.8 pp, medium confidence
49GPT-5.6 Luna50.8%estimated ± 6.8 pp, medium confidence
50Qwen3.8-Flash-Next50.8%estimated ± 6.8 pp, medium confidence
51GPT-5.3 Codex50.7%estimated ± 6.8 pp, medium confidence
52Grok 4.350.5%estimated ± 6.8 pp, medium confidence
53Qwen3.6 Plus50.5%estimated ± 6.8 pp, medium confidence
54Qwen3.5 397B50.4%estimated ± 6.8 pp, medium confidence
55GPT-5.2-Codex50.3%estimated ± 6.8 pp, low confidence
56Qwen3.8-27B50.3%estimated ± 6.8 pp, low confidence
57GPT-5.150.3%estimated ± 6.8 pp, low confidence
58Claude Opus 4.6 (Adaptive)50.3%estimated ± 6.8 pp, low confidence
59Kimi K2.550.3%estimated ± 6.8 pp, low confidence
60Step 3.7 Flash50.3%estimated ± 6.8 pp, low confidence
61Qwen3.5-122B-A10B50.3%estimated ± 6.8 pp, low confidence
62Qwen3.5-27B50.3%estimated ± 6.8 pp, low confidence
63Qwen3.6-35B-A3B50.3%estimated ± 6.8 pp, low confidence
64Gemini 2.5 Pro50.3%estimated ± 6.8 pp, low confidence
65Qwen3.6-27B50.3%estimated ± 6.8 pp, low confidence
66Muse Glimmer 30B50.3%estimated ± 6.8 pp, low confidence
67GPT-550.3%estimated ± 6.8 pp, low confidence
68Claude Opus 4.5 Thinking50.3%estimated ± 6.8 pp, low confidence
69Inkling-Small50.3%estimated ± 6.8 pp, low confidence
70Inkling50.3%estimated ± 6.8 pp, low confidence
71Gemma 4 31B50.3%estimated ± 6.8 pp, low confidence
72GPT-5.4 mini50.3%estimated ± 6.8 pp, low confidence
73MiMo-V2.6-Flash50.3%estimated ± 6.8 pp, low confidence
74GLM-5V-Turbo50.3%estimated ± 6.8 pp, low confidence
75Qwen3.5-35B-A3B50.3%estimated ± 6.8 pp, low confidence
76Claude Opus 4.650.3%estimated ± 6.8 pp, low confidence
77GPT-5.1-Codex50.3%estimated ± 6.8 pp, low confidence
78GPT-5.1-Codex-Max50.3%estimated ± 6.8 pp, low confidence
79Claude Opus 4.550.3%estimated ± 6.8 pp, low confidence
80Claude Sonnet 4.650.3%estimated ± 6.8 pp, low confidence
81o350.3%estimated ± 6.8 pp, low confidence
82MiMo-V2-Omni50.3%estimated ± 6.8 pp, low confidence
83Gemma 4 12B50.3%estimated ± 6.8 pp, low confidence
84Gemma 4 26B A4B50.3%estimated ± 6.8 pp, low confidence
85o4-mini50.3%estimated ± 6.8 pp, low confidence
86Grok 450.3%estimated ± 6.8 pp, low confidence
87Gemini 2.5 Flash50.3%estimated ± 6.8 pp, low confidence
88GPT-5.4 nano50.3%estimated ± 6.8 pp, low confidence
89Qwen3.5-4B50.3%estimated ± 6.8 pp, low confidence
90Claude Haiku 4.5 Thinking48.5%estimated ± 7.0 pp, low confidence
91Claude Haiku 4.547.1%estimated ± 7.0 pp, low confidence
92Nemotron 3 Nano Omni 30B A3B46.5%estimated ± 7.0 pp, low confidence
93MiniMax M345.1%measured
94Claude Opus 4.743.6%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General