Inference
EN
LiveBench

LiveBench Instruction Following

Mean score over paraphrasing, simplifying, summarising and story generation, each marked on whether the stated constraints were actually met. Not in LiveBench 2.0.

ModelsModels with a result.66
Top resultBest result on this benchmark.81.4%Gemini 3.8 Flash
Top-3 spreadPoints from first to third.2.3 pts
UpdatedDate the source changed the results.25 Jun 2026

Result and cost

65 of 66 results

Results

66 results
#ModelResultThe score from the source.CostUS dollars to run one task.
01Gemini 3.8 Flashhigh effortAbove fitFar above this model's index.81.4%$0.169
02Gemini 3.7 Flashhigh effortAbove fitFar above this model's index.79.9%$0.097
03Gemini 3.1 Pro Previewhigh effortAbove fitFar above this model's index.79.1%$0.139
04Muse Spark 1.3xhigh effort78.0%$0.133
05Qwen3.8-Flash-NextAbove fitFar above this model's index.77.1%$0.018
06Claude Fable 5max effort75.8%$0.924
07Gemini 3.5 Flashhigh effort75.6%$0.133
07GPT-6 Astramax effort75.6%$1.24
09Gemini 3.6 Flashhigh effort75.4%$0.115
10Grok 4.7xhigh effort75.3%$0.237
11Muse Spark 1.2xhigh effort74.3%$0.072
12GPT-6.1 Solmax effort74.2%$0.24
13Qwen3.8 Max74.1%$0.082
14Qwen3.7 Max74.0%$0.079
15Smaug Mini73.9%$0.014
16Muse Wondermax effort73.0%$0.329
16Claude Fable 5.1max effort73.0%$1.20
18Qwen3.8-27B72.7%$0.013
19Claude Opus 4.8max effort72.0%$0.471
20Grok 4.671.9%$0.044
21GPT-5.6 Solmax effort71.8%$0.418
22Grok 4.571.5%$0.047
23Kimi K371.4%$0.118
24Smaug Agentic71.0%$0.105
24DeepSeek V4 Flash Vision Exp71.0%$0.036
26GPT-5.5xhigh effort70.7%$0.227
27GPT-5.4xhigh effort70.2%$0.259
28Inklingxhigh effort70.1%$0.067
29DeepSeek V4.1 Flash 1017max effort70.0%$0.016
30Muse Spark 1.1xhigh effort69.6%$0.047
31GLM-5.369.3%$0.271
31Smaug Flash69.3%$0.0073
33GPT-6 Solmax effort68.6%$0.419
34DeepSeek V4 Pro 081367.7%$0.021
35Gemini 3.5 Flash-Litehigh effort67.2%$0.023
35GPT-5.4 nanoxhigh effort67.2%$0.045
37Claude Opus 4.7xhigh effort66.7%$0.281
38GPT-5.2-Codex66.4%$0.071
39Claude Opus 5.5max effort65.7%$0.937
40DeepSeek V4 Flash 073165.5%$0.0045
41Grok Build 0.165.2%$0.0009
42Mistral Large 4high effort64.8%$0.038
43GPT-5.6 Terramax effort64.6%$0.332
44Kimi K2.6 Thinking64.4%$0.067
45Claude Sonnet 5xhigh effort63.9%$0.259
46Claude Opus 5max effort63.8%$0.328
47Claude Opus 4.6 Thinking Autohigh effort63.3%$0.335
48Claude Sonnet 4.6 Thinking Automedium effort63.2%$0.248
49DeepSeek V4 Flash 042363.1%$0.0073
50Grok 4.362.8%$0.027
51Claude Opus 4.5 Thinking 64Khigh effort62.5%$0.40
52DeepSeek V4 Pro 042362.4%$0.02
53GLM-5.262.3%$0.062
54GPT-5.2high effort61.8%$0.102
55Ox Alpha Max60.3%$0
56GPT-5.6 Lunamax effort60.1%$0.153
57GPT-5.4 minixhigh effort59.8%$0.172
58Qwen3.6 Plus58.3%$0.05
59MiniMax M357.5%$0.019
60Claude Sonnet 5.5max effort56.8%$0.914
61Kimi K2.7 Code56.3%$0.047
62GPT-6 Lunamax effort55.9%$0.035
63Nemotron 3.5 Lightning 30B A3B54.9%$0.012
64Claude Haiku 5.5max effort54.7%$0.043
65Qwen3.6-27B53.2%$0.046
66GLM-5.3-Flash52.8%$0.019