Preservation

Subcategory of Text Editing. 116 models scored.

Model Leaderboard

All models ranked by their Preservation subcategory score.

# Model Preservation Text Editing Overall
1 Claude Opus 4.6 100.00% 98.35% 92.35%
2 Claude Opus 4.5 100.00% 97.69% 89.69%
3 Claude Sonnet 4.6 99.82% 96.37% 91.15%
4 Claude Sonnet 4 99.69% 99.13% 88.72%
5 Claude Haiku 4.5 99.64% 96.81% 85.14%
6 Qwen 3.5 Plus (2026-02-15) 99.38% 98.10% 85.96%
7 Claude Sonnet 4.5 99.31% 99.02% 88.03%
8 Grok 4 99.23% 98.76% 88.12%
9 Claude Sonnet 4.6 (Reasoning) 99.06% 98.30% 93.66%
10 Gemini 3.1 Pro (Preview) 99.03% 98.51% 94.37%
11 Gemini 3 Flash (Preview) 98.90% 97.54% 85.35%
12 Gemini 2.5 Pro 98.80% 98.58% 88.53%
13 Grok 4.20 (Beta, Reasoning) 98.80% 98.69% 91.49%
14 Claude Opus 4.6 (Reasoning) 98.72% 98.86% 95.02%
15 Z.AI GLM 4.7 98.67% 98.22% 88.69%
16 Gemini 2.5 Flash 98.67% 97.83% 80.60%
17 Z.AI GLM 5 98.62% 98.59% 91.23%
18 Grok 4 Fast 98.60% 97.26% 86.15%
19 GPT-5 98.57% 98.90% 91.93%
20 Qwen 3.5 27B 98.55% 98.69% 90.85%
21 Gemini 3 Pro (Preview) 98.47% 98.86% 88.79%
22 Claude Opus 4 98.34% 97.25% 87.69%
23 Z.AI GLM 4.6 98.31% 97.78% 89.11%
24 Qwen 3.5 122B 98.28% 96.31% 91.53%
25 MoonshotAI: Kimi K2.5 98.19% 97.79% 91.04%
26 Gemini 3 Flash (Preview, Reasoning) 98.19% 98.12% 90.50%
27 GPT-5.1 98.18% 98.54% 92.54%
28 GPT-4o, May 13th (temp=0) 98.16% 95.35% 85.36%
29 Gemini 3.1 Flash Lite (Preview) 98.12% 96.46% 85.87%
30 MiniMax M2.5 98.01% 96.02% 88.71%
31 Stealth: Healer Alpha 97.86% 96.04% 85.93%
32 Z.AI GLM 5 Turbo 97.83% 98.17% 94.27%
33 Claude 3.7 Sonnet 97.73% 97.12% 83.39%
34 GPT-4.1 Mini 97.60% 95.62% 83.20%
35 ByteDance Seed 1.6 97.58% 98.40% 90.70%
36 ByteDance Seed 2.0 Lite 97.58% 95.03% 84.80%
37 GPT-4o, May 13th (temp=1) 97.54% 92.41% 83.80%
38 GPT-5 Mini 97.51% 97.13% 92.62%
39 Hermes 3 405B 97.48% 89.14% 82.86%
40 GPT-5.4 (Reasoning) 97.42% 98.42% 93.24%
41 Gemini 2.5 Flash (Reasoning) 97.39% 98.12% 86.51%
42 Grok 4.1 Fast 97.29% 97.87% 89.55%
43 GPT-5.4 (Reasoning, Low) 97.21% 98.01% 91.41%
44 Aion 2.0 97.21% 95.34% 89.21%
45 o4 Mini High 97.09% 94.36% 90.29%
46 Mistral Small Creative 96.98% 90.31% 73.27%
47 Claude 3.5 Sonnet 96.96% 96.57% 84.24%
48 GPT-4o, Aug. 6th (temp=0) 96.79% 93.77% 82.45%
49 GPT-4.1 96.66% 94.40% 88.68%
50 Mistral Medium 3.1 96.56% 93.77% 77.83%
51 Stealth: Hunter Alpha 96.45% 95.53% 87.34%
52 Qwen 3.5 397B A17B 96.38% 98.05% 91.73%
53 GPT-5.2 96.38% 97.54% 90.26%
54 o4 Mini 96.24% 90.61% 88.35%
55 GPT-5.4 96.10% 96.73% 84.32%
56 MiniMax M2.7 95.84% 92.14% 89.10%
57 Qwen 2.5 72B 95.78% 89.18% 75.46%
58 Qwen3 235B A22B Instruct 2507 95.60% 91.75% 80.10%
59 GPT-5.4 Mini (Reasoning) 95.50% 95.78% 90.65%
60 Gemma 3 27B 95.49% 86.63% 77.85%
61 ByteDance Seed 1.6 Flash 95.32% 91.64% 73.27%
62 GPT-4o, Aug. 6th (temp=1) 95.01% 86.72% 82.62%
63 Llama 3.1 70B 94.67% 92.10% 78.40%
64 Mistral Large 94.67% 95.14% 80.15%
65 Grok 4.20 (Beta) 94.41% 95.49% 83.85%
66 Gemma 3 4B 94.35% 78.38% 68.57%
67 Writer: Palmyra X5 94.28% 91.20% 79.57%
68 Qwen 3.5 35B 94.21% 94.95% 88.00%
69 Qwen 3 32B 93.83% 89.95% 82.21%
70 Mistral Small 4 (Reasoning) 93.47% 90.58% 82.39%
71 Llama 3.1 Nemotron 70B 93.45% 87.26% 74.70%
72 Qwen 3.5 Flash 93.04% 92.80% 86.38%
73 DeepSeek V3 (2024-12-26) 92.60% 93.58% 83.68%
74 Qwen 3.5 9B 92.48% 85.35% 86.05%
75 Mistral Small 3.2 24B 92.38% 89.48% 78.60%
76 Gemini 2.5 Flash Lite 92.17% 92.13% 81.08%
77 GPT-5.4 Mini (Reasoning, Low) 91.94% 92.63% 85.75%
78 GPT-5 Nano 91.91% 82.74% 82.60%
79 DeepSeek V3.2 91.71% 95.78% 82.25%
80 Z.AI GLM 4.5 91.69% 95.32% 86.27%
81 Mistral Large 3 91.61% 94.09% 85.43%
82 Mistral Large 2 91.61% 94.16% 82.41%
83 DeepSeek-V2 Chat 91.53% 90.90% 84.83%
84 Mistral Small 4 91.45% 91.00% 76.46%
85 Inception Mercury 2 91.31% 85.26% 83.85%
86 DeepSeek V3 (2025-03-24) 90.56% 89.57% 81.99%
87 Z.AI GLM 4.7 Flash 90.40% 85.82% 84.82%
88 GPT-5.4 Nano (Reasoning) 90.10% 83.32% 81.36%
89 ByteDance Seed 2.0 Mini 90.10% 91.08% 86.91%
90 WizardLM 2 8x22b 89.67% 88.13% 71.07%
91 Nemotron 3 Super 89.49% 86.34% 84.56%
92 Ministral 3 14B 89.12% 86.20% 72.54%
93 Gemini 2.5 Flash Lite (Reasoning) 89.04% 94.54% 85.75%
94 GPT-5.4 Mini 88.47% 90.60% 82.43%
95 Gemma 3 12B 87.66% 85.18% 78.41%
96 Inception Mercury 87.15% 79.53% 79.50%
97 GPT-4o Mini (temp=1) 86.94% 85.78% 79.08%
98 GPT-4o Mini (temp=0) 86.87% 84.62% 78.29%
99 Arcee AI: Trinity Mini 85.07% 73.88% 70.90%
100 GPT-5.4 Nano (Reasoning, Low) 85.00% 82.23% 79.48%
101 DeepSeek V3.1 84.70% 87.27% 82.39%
102 GPT-5.4 Nano 84.23% 79.22% 74.40%
103 Ministral 8B 83.69% 77.52% 64.87%
104 Ministral 3 8B 83.53% 78.52% 71.76%
105 Arcee AI: Trinity Large (Preview) 83.17% 86.62% 73.33%
106 Mistral NeMO 82.35% 73.69% 65.04%
107 Llama 3.1 8B 80.76% 75.45% 63.37%
108 Nemotron 3 Nano 78.40% 75.81% 77.73%
109 GPT-4.1 Nano 78.02% 76.06% 71.94%
110 Ministral 3B 74.55% 70.91% 61.29%
111 Ministral 3 3B 73.12% 69.80% 67.22%
112 Cohere Command R+ (Aug. 2024) 70.67% 68.40% 69.03%
113 Hermes 3 70B 66.87% 63.34% 72.57%
114 Rocinante 12B 63.96% 56.31% 54.55%
115 Claude 3 Haiku 63.45% 64.36% 71.19%
116 LFM2 24B 57.43% 71.56% 58.77%