Skip to content

Benchmark

Arena ELO - Instruction Following

Instruction following benchmarks measure how closely models follow instructions. Higher scores are better.

Models measured
366
Scores recorded
423
Sources
1
IQ weight
1.18

Score distribution

1,294.5
Mean
1,332.7
Median
155.5
Std dev
1,056.7
10th percentile
1,465
90th percentile
780.7
Min
780.71,539.1
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.18 for chat models, measured over 366 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51GPT-5.6 TerraxHigh· LMArena data1,454.9
52GPT 5.4· LMArena data1,454.8
53GPT-6 AstraMax· LMArena data1,454.8
54Claude Sonnet 5High· LMArena data1,452.7
55GLM 5.1· LMArena data1,452
56Kimi K2.6· LMArena data1,450.7
57DeepSeek V4 ProHigh, 20260813· LMArena data1,447.6
58Step 5 Preview· LMArena data1,445.9
59Grok 4.5· LMArena data1,445.6
60DeepSeek V4 Pro· LMArena data1,445.5
61MiMo V2 Pro· LMArena data1,445.5
62MiMo V2.6 Flash· LMArena data1,444.4
63Muse Spark· LMArena data1,442.9
64GPT 5.1High· LMArena data1,442.2
65GPT-6.1 TerraMax· LMArena data1,441.5
66Qwen3.7 Plus· LMArena data1,438.8
67Qwen3.6 Max Preview· LMArena data1,438.6
68Gemini 2.5 Pro· LMArena data1,436.6
69GPT-5.6 LunaxHigh· LMArena data1,436.5
70Gemini 3 Flash· LMArena data1,436.4
71DeepSeek-V4-Pro PreviewHigh· LMArena data1,435.8
72Claude Opus 4.1Thinking 16K, 20250805· LMArena data1,435.5
73MiMo V2.5· LMArena data1,434.4
74GPT-6.2 SolMax· LMArena data1,434.1
75Claude Opus 4.120250805· LMArena data1,433.3
76Qwen 3.8 27B· LMArena data1,433.2
77Gemma 4 31B IT· LMArena data1,432.4
78Grok 4.6High· LMArena data1,431.3
79Kimi K2.5Thinking· LMArena data1,430.5
80Kimi K2.5Instant· LMArena data1,429.8
81MiMo V2 Omni· LMArena data1,428.5
82GLM 5· LMArena data1,428.2
83Inkling FP4· LMArena data1,427.4
84nova-experimental-chat26-02-10· LMArena data1,426.7
85Hy3 (Deep Infra)· LMArena data1,425.9
86MiniMax M3· LMArena data1,425.2
87Beam 1.0 501B A23B· LMArena data1,424.3
88Qwen3.6 Plus· LMArena data1,424.1
89Qwen3.5 397B A17B· LMArena data1,423.9
90GPT-6.2 AstraMax· LMArena data1,423
91Grok 4.20Reasoning, 0309, beta· LMArena data1,420.9
92Gemma 4 26B A4B· LMArena data1,420.6
93DeepSeek V4 Flash· LMArena data1,420.6
94Grok 4.20 Multi-Agent0309, beta· LMArena data1,420.5
95Gemini 3.5 Flash Lite· LMArena data1,420.1
96Qwen3 Max Preview· LMArena data1,419.7
97Nemotron 3 UltraNVFP4· LMArena data1,418
98GPT 5.1· LMArena data1,418
99DeepSeek-V4-Flash PreviewHigh· LMArena data1,417.7
100GPT-5.2 Chat20260210· LMArena data1,416.3

Showing 51–100 of 423 results.

Data credits: LMArena, CC BY 4.0 (licence)