Inference
EN
LiveBench

LiveBench Agentic Coding

Mean score over agentic tasks in live JavaScript, TypeScript and Python repositories, where the model changes a codebase rather than writing one function. Not in LiveBench 2.0.

ModelsModels with a result.66
Top resultBest result on this benchmark.77.3%DeepSeek V4.1 Flash 1017
Top-3 spreadPoints from first to third.11.2 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.
01DeepSeek V4.1 Flash 1017max effortAbove fitFar above this model's index.77.3%$0.064
02Claude Opus 5.5max effort71.7%$4.48
03Claude Fable 5.1max effort66.1%$6.22
04Claude Opus 5max effort65.2%$5.34
05DeepSeek V4 Flash Vision Exp65.1%$0.109
06Qwen3.8 Max64.6%$1.75
06Smaug Agentic64.6%$1.81
08Muse Spark 1.3xhigh effort64.1%$1.17
09Muse Wondermax effort62.7%$4.45
10Claude Fable 5max effort62.2%$8.86
10Kimi K362.2%$1.76
12Qwen3.8-Flash-Next61.6%$0.154
13Qwen3.8-27B61.4%$0.947
14Smaug Flash61.1%$0.078
15GLM-5.360.9%$1.60
16Smaug Mini60.8%$1.07
17Claude Sonnet 5xhigh effort59.4%$2.14
18Muse Spark 1.1xhigh effort58.5%$1.73
19Gemini 3.7 Flashhigh effort58.3%$0.833
20Muse Spark 1.2xhigh effort57.6%$3.92
21GPT-6 Astramax effort57.3%$2.43
22Claude Haiku 5.5max effort57.2%$0.211
22Mistral Large 4high effort57.2%$0.91
24Grok 4.657.0%$1.79
25GLM-5.3-Flash56.8%$0.111
26Grok 4.556.5%$0.684
27Claude Sonnet 5.5max effort56.3%$5.57
28GPT-5.6 Solmax effort56.2%$3.40
29GPT-5.6 Terramax effort54.9%$1.35
29DeepSeek V4 Pro 081354.9%$0.084
31GPT-6.1 Solmax effort54.5%$0.405
32Gemini 3.8 Flashhigh effort54.2%$1.58
33Grok 4.7xhigh effort54.0%$4.15
33GPT-5.5xhigh effort54.0%$2.18
35GPT-5.4xhigh effort53.8%$1.73
36GPT-6 Solmax effort52.9%$1.06
37Ox Alpha Max52.6%$0
38GLM-5.251.8%$1.46
39GPT-6 Lunamax effort51.2%$0.032
40Claude Opus 4.7xhigh effort50.7%$2.64
41Claude Opus 4.8max effort50.5%$4.96
42GPT-5.2high effort50.3%$1.42
43GPT-5.2-Codex49.4%$0.594
43Inklingxhigh effort49.4%$2.23
45Gemini 3.5 Flashhigh effort49.0%$1.02
45Claude Opus 4.6 Thinking Autohigh effort49.0%$1.24
47GPT-5.6 Lunamax effort48.4%$0.494
48Kimi K2.6 Thinking46.9%$0.717
49DeepSeek V4 Flash 073146.8%$0.693
49GPT-5.4 nanoxhigh effort46.8%$0.529
51Grok Build 0.145.8%$0.258
52Kimi K2.7 Code45.7%$0.395
53Gemini 3.5 Flash-Litehigh effort45.3%$0.33
54Gemini 3.1 Pro Previewhigh effort44.1%$1.31
55Qwen3.7 Max43.6%$0.796
56Gemini 3.6 Flashhigh effort43.4%$1.35
57DeepSeek V4 Pro 042342.6%$0.142
57Claude Sonnet 4.6 Thinking Automedium effort42.6%$0.63
59GPT-5.4 minixhigh effort41.7%$1.18
60Qwen3.6 Plus41.4%$1.55
61MiniMax M340.7%$0.424
62Claude Opus 4.5 Thinking 64Khigh effort39.7%$1.42
63Qwen3.6-27B39.3%$0.813
64DeepSeek V4 Flash 042337.6%$0.032
65Nemotron 3.5 Lightning 30B A3B19.9%$0.098
66Grok 4.318.5%$0.167