Home/Benchmarks/LiveBench Design

LiveBench Design

Document generation (DOCX/PPT/PDF) · 22 models · source LiveBench · updated 2026-10

AA Intelligence Index v5GPQA DiamondTerminal-Bench 4.0Vals IndexAA Intelligence Index v4.3.2
What this benchmark measures

The model must produce Word, PowerPoint and PDF files that actually open, scored on structural integrity, formatting correctness and content organisation.

Why it gets its own table

The real demand of office work. Interestingly small models do unusually well here: Haiku 5.5 (83.1) and DeepSeek V4.1 Flash (83.4) almost match Opus 5.5 (86.9).

Full ranking

#ModelScoreVs. bestInput $Output $Type
1 Claude Opus 5.5Anthropic 86.9 Best overall $4.00 $20.00
2 Claude Sonnet 5.5Anthropic 85.0 98% $2.00 $10.00
3 Claude Fable 5.1Anthropic 83.4 96% $10.00 $50.00
4 DeepSeek V4.1 Flash Open weightsDeepSeek 83.4 96% $0.30 $1.20 Open weights
5 Claude Haiku 5.5Anthropic 83.1 96% $0.10 $0.50
6 GPT-6 LunaOpenAI 82.5 95% $0.10 $0.50
7 GPT-6.1 TerraOpenAI 82.3 95% $1.00 $5.00
8 GPT-6.1 SolOpenAI 82.1 94% $2.00 $10.00
9 GPT-6 AstraOpenAI 82.0 94% $10.00 $50.00
10 Qwen3.8 MaxAlibaba Qwen 78.6 90% $2.00 $8.00
11 MiMo-V2.6 Pro Open weightsXiaomi 78.5 90% $0.35 $1.40 Open weights
12 Grok 4.7xAI 78.0 90% $2.00 $6.00
13 Kimi K3.1 Pro Open weightsMoonshot AI 77.8 90% $2.00 $10.00 Open weights
14 Kimi K3 Open weightsMoonshot AI 77.2 89% $2.00 $15.00 Open weights
15 GLM-5.3 Flash Open weightsZ.ai 77.2 89% $0.20 $0.80 Open weights
16 Step 5 PreviewStepFun 76.9 88% $0.50 $2.00
17 Qwen3.8 27B Open weightsAlibaba Qwen 75.9 87% $0.10 $0.40 Open weights
18 MiniMax M3.1 FlashMiniMax 74.0 85% $0.15 $0.60
19 Gemini 4 ArgonGoogle 73.7 85% $2.00 $10.00
20 Gemini 3.8 FlashGoogle 72.7 84% $0.75 $3.75
21 Nemotron 3 Ultra Open weightsNVIDIA 62.8 72% $0.40 $1.60 Open weights
22 Gemma 4 31B Open weightsGoogle 58.4 67% $0.10 $0.40 Open weights