Constraint Adherence

Subcategory of Rule Following. 155 models scored.

Model Leaderboard

All models ranked by their Constraint Adherence subcategory score.

# Model Constraint Adherence Rule Following Overall
1 Qwen3.7 Max 95.76% 95.76% 95.75%
2 Gemini 3.5 Flash (Reasoning) 92.04% 92.04% 94.08%
3 Gemini 3.1 Pro (Preview) 91.21% 91.21% 94.37%
4 Claude Opus 4.6 (Reasoning) 89.78% 89.78% 95.02%
5 Z.AI GLM 5.1 88.41% 88.41% 94.37%
6 Z.AI GLM 5 Turbo 86.78% 86.78% 94.27%
7 Claude Sonnet 4.6 (Reasoning) 85.73% 85.73% 93.66%
8 Gemma 4 31B (Reasoning) 85.00% 85.00% 91.71%
9 Claude Opus 4.6 83.11% 83.11% 92.35%
10 Grok 4.3 (Reasoning) 82.80% 82.80% 93.60%
11 Qwen3.6 Max Preview 82.79% 82.79% 94.54%
12 Claude Sonnet 4.6 82.50% 82.50% 91.15%
13 Grok 4.20 (Reasoning) 82.04% 82.04% 91.39%
14 Claude Sonnet 4 81.52% 81.52% 88.72%
15 Qwen 3.5 122B 80.00% 80.00% 91.53%
16 GPT-5.5 (Reasoning) 79.40% 79.40% 92.98%
17 Qwen 3.5 397B A17B 79.39% 79.39% 91.73%
18 GPT-5.4 (Reasoning) 79.29% 79.29% 93.24%
19 ByteDance Seed 1.6 77.71% 77.71% 90.70%
20 Qwen 3.6 35B 77.34% 77.34% 89.05%
21 GPT-5 77.13% 77.13% 91.93%
22 MoonshotAI: Kimi K2.6 76.94% 76.94% 92.31%
23 GPT-5.5 (Reasoning, Low) 76.90% 76.90% 92.59%
24 Claude Sonnet 4.5 76.80% 76.80% 88.03%
25 GPT-5 Mini 76.44% 76.44% 92.62%
26 Qwen 3.5 27B 76.04% 76.04% 90.85%
27 Grok 4.20 (Beta, Reasoning) 75.31% 75.31% 91.49%
28 Gemma 4 26B (Reasoning) 74.75% 74.75% 91.49%
29 Gemini 3 Flash (Preview, Reasoning) 74.48% 74.48% 90.50%
30 GPT-4o, Aug. 6th (temp=0) 74.19% 74.19% 82.45%
31 GPT-5.1 74.05% 74.05% 92.54%
32 Claude Opus 4.7 (Reasoning) 74.04% 74.04% 93.23%
33 Claude 3.7 Sonnet 73.78% 73.78% 83.39%
34 GPT-4o, May 13th (temp=0) 73.24% 73.24% 85.36%
35 DeepSeek V4 Pro (Reasoning) 72.74% 72.74% 90.10%
36 Gemma 4 31B 72.72% 72.72% 86.91%
37 o4 Mini High 72.70% 72.70% 90.29%
38 Claude Opus 4.5 72.61% 72.61% 89.69%
39 GPT-5.5 72.34% 72.34% 89.09%
40 MoonshotAI: Kimi K2.5 72.03% 72.03% 91.04%
41 Gemma 4 26B 71.75% 71.75% 85.84%
42 Qwen 3.6 Flash 71.50% 71.50% 90.65%
43 Qwen 3.6 27B 71.37% 71.37% 89.72%
44 Grok 4.1 Fast 70.87% 70.87% 89.55%
45 Claude Opus 4.8 (Reasoning, Low) 70.56% 70.56% 92.14%
46 Claude Opus 4 70.37% 70.37% 87.69%
47 Claude Haiku 4.5 70.35% 70.35% 85.14%
48 Claude Opus 4.8 (Reasoning) 70.27% 70.27% 92.22%
49 GPT-5.4 (Reasoning, Low) 70.02% 70.02% 91.41%
50 GPT-4o, May 13th (temp=1) 69.88% 69.88% 83.80%
51 Claude 3.5 Sonnet 69.67% 69.67% 84.24%
52 MiniMax M3 69.18% 69.18% 90.88%
53 Z.AI GLM 4.7 69.16% 69.16% 88.69%
54 MiniMax M2.7 68.90% 68.90% 89.10%
55 DeepSeek-V2 Chat 68.78% 68.78% 84.83%
56 Claude Opus 4.7 68.08% 68.08% 89.93%
57 DeepSeek V3 (2025-03-24) 67.94% 67.94% 81.99%
58 Grok 4 Fast 67.91% 67.91% 86.15%
59 GPT-4o, Aug. 6th (temp=1) 67.91% 67.91% 82.62%
60 Z.AI GLM 5 67.78% 67.78% 91.23%
61 Qwen 3.5 Plus (2026-04-20) 67.53% 67.53% 91.51%
62 Qwen 3.5 35B 67.42% 67.42% 88.00%
63 Writer: Palmyra X5 67.19% 67.19% 79.57%
64 GPT-5.2 67.10% 67.10% 90.26%
65 Gemini 2.5 Flash Lite (Reasoning) 66.81% 66.81% 85.75%
66 GPT-4.1 66.78% 66.78% 88.68%
67 DeepSeek V3 (2024-12-26) 66.39% 66.39% 83.68%
68 DeepSeek V3.1 66.15% 66.15% 82.39%
69 Z.AI GLM 4.6 65.85% 65.85% 89.11%
70 Z.AI GLM 4.7 Flash 65.63% 65.63% 84.82%
71 Qwen3 235B A22B Instruct 2507 65.42% 65.42% 80.10%
72 Gemini 3 Flash (Preview) 65.14% 65.14% 85.35%
73 o4 Mini 64.61% 64.61% 88.35%
74 DeepSeek V4 Flash (Reasoning) 64.50% 64.50% 89.01%
75 Gemini 3 Pro (Preview) 64.47% 64.47% 88.79%
76 Mistral Large 3 64.41% 64.41% 85.43%
77 Xiaomi MIMO v2.5 Pro 64.29% 64.29% 87.36%
78 Qwen 3.5 Plus (2026-02-15) 64.21% 64.21% 85.96%
79 Mistral Small 3.2 24B 64.08% 64.08% 78.58%
80 Z.AI GLM 4.5 63.79% 63.79% 86.27%
81 Aion 2.0 63.77% 63.77% 89.21%
82 DeepSeek V4 Pro 63.74% 63.74% 82.63%
83 Stealth: Hunter Alpha 63.63% 63.63% 87.34%
84 Llama 3.1 70B 63.45% 63.45% 78.40%
85 Qwen 3.5 Flash 63.19% 63.19% 86.38%
86 Grok 4 63.09% 63.09% 88.12%
87 Mistral Large 2 63.05% 63.05% 82.41%
88 MiniMax M2.5 62.69% 62.69% 88.71%
89 Gemini 3.1 Flash Lite (Reasoning) 62.26% 62.26% 86.41%
90 Mistral Small 4 62.17% 62.17% 76.46%
91 Gemini 3.1 Flash Lite 61.21% 61.21% 85.75%
92 Gemma 3 12B 61.05% 61.05% 78.41%
93 Qwen 3.5 9B 60.98% 60.98% 86.05%
94 Gemini 2.5 Pro 60.89% 60.89% 88.53%
95 Xiaomi MIMO v2.5 60.75% 60.75% 85.05%
96 Gemini 3.5 Flash (Reasoning, Minimal) 60.38% 60.38% 86.47%
97 Mistral Small 4 (Reasoning) 60.28% 60.28% 82.39%
98 Gemini 2.5 Flash (Reasoning) 59.97% 59.97% 86.51%
99 Gemini 2.5 Flash Lite 59.96% 59.96% 81.08%
100 Grok 4.20 59.71% 59.71% 81.70%
101 Hermes 3 405B 59.17% 59.17% 82.86%
102 Gemini 3.1 Flash Lite (Preview) 59.04% 59.04% 85.87%
103 GPT-4o Mini (temp=0) 58.84% 58.84% 78.29%
104 ByteDance Seed 2.0 Mini 58.77% 58.77% 86.91%
105 Cohere Command R+ (Aug. 2024) 58.70% 58.70% 69.03%
106 GPT-4.1 Mini 58.59% 58.59% 83.20%
107 GPT-5.4 58.11% 58.11% 84.32%
108 GPT-5 Nano 57.57% 57.57% 82.60%
109 Gemini 2.5 Flash 57.47% 57.47% 80.60%
110 Nemotron 3 Super 57.43% 57.43% 84.56%
111 GPT-5.4 Mini (Reasoning) 57.38% 57.38% 90.65%
112 DeepSeek V4 Flash 57.32% 57.32% 82.02%
113 GPT-4o Mini (temp=1) 56.50% 56.50% 79.08%
114 Stealth: Healer Alpha 56.03% 56.03% 85.93%
115 GPT-OSS 120B 55.03% 55.03% 86.44%
116 Inception Mercury 2 54.41% 54.41% 83.85%
117 Grok 4.20 (Beta) 53.89% 53.89% 83.85%
118 DeepSeek V3.2 53.75% 53.75% 82.25%
119 Hermes 3 70B 53.00% 53.00% 72.57%
120 Claude 3 Haiku 51.15% 51.15% 71.19%
121 Ministral 3 14B 50.83% 50.83% 72.54%
122 Llama 3.1 Nemotron 70B 50.62% 50.62% 74.70%
123 Cydonia 24B V4.1 50.36% 50.36% 75.09%
124 Mistral Large 49.87% 49.87% 80.15%
125 Grok 4.3 49.02% 49.02% 78.66%
126 Mistral Medium 3.1 48.60% 48.60% 77.83%
127 Mistral Small Creative 48.15% 48.15% 73.27%
128 Gemma 3 27B 47.98% 47.98% 77.85%
129 ByteDance Seed 1.6 Flash 47.15% 47.15% 73.27%
130 Qwen 3 32B 46.83% 46.83% 82.21%
131 GPT-5.4 Mini 46.32% 46.32% 82.43%
132 Stealth: Aurora Alpha 44.19% 44.19% 83.79%
133 Z.AI GLM 4.5 Air 44.11% 44.11% 83.12%
134 Nemotron 3 Nano 43.47% 43.47% 77.73%
135 Rocinante 12B 41.51% 41.51% 54.54%
136 Skyfall 36B V2 41.44% 41.44% 65.76%
137 GPT-4.1 Nano 40.88% 40.88% 71.94%
138 Inception Mercury 39.68% 39.68% 79.50%
139 Arcee AI: Trinity Large (Preview) 38.52% 38.52% 73.33%
140 ByteDance Seed 2.0 Lite 36.85% 36.85% 84.80%
141 Mistral NeMO 34.11% 34.11% 65.04%
142 Llama 3.1 8B 34.03% 34.03% 63.35%
143 GPT-5.4 Mini (Reasoning, Low) 33.99% 33.99% 85.75%
144 GPT-5.4 Nano (Reasoning, Low) 31.65% 31.65% 79.48%
145 Qwen 2.5 72B 31.55% 31.55% 75.46%
146 Ministral 3 8B 31.34% 31.34% 71.76%
147 WizardLM 2 8x22b 28.27% 28.27% 71.06%
148 GPT-5.4 Nano (Reasoning) 27.15% 27.15% 81.36%
149 Gemma 3 4B 26.37% 26.37% 68.57%
150 Ministral 3B 24.45% 24.45% 61.29%
151 LFM2 24B 24.12% 24.12% 58.77%
152 Arcee AI: Trinity Mini 23.57% 23.57% 70.90%
153 GPT-5.4 Nano 20.94% 20.94% 74.40%
154 Ministral 3 3B 15.87% 15.87% 67.22%
155 Ministral 8B 15.27% 15.27% 64.87%