BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12

LiveBench Agentic Coding

Agentic coding category of LiveBench 1.0, frozen at its final release (25 June 2026). LiveBench 2.0 dropped it.

As of 24 Oct 2026, DeepSeek V4.1 Flash leads LiveBench Agentic Coding on BenchLeader with 77.3%, ahead of Claude Opus 5.5 at 71.7%, across 67 model configurations with a published result.

Published by
LiveBench
Category
Agents & tools
Index weight
Reference only
Models
67
Data as of
24 Oct 2026
Newest result
25 Jun 2026

The publisher has posted no new results on this board since 25 Jun 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

LiveBench (livebench.ai), White et al. 2024.

What the test looks like

LiveBench's agentic coding category: the model works through multi-step coding tasks with tool use in an agent loop, on freshly written problems.

How it is scored

Percent of tasks completed, from LiveBench 1.0's final release (25 June 2026). LiveBench 2.0 dropped the category, so these scores no longer update.

What to keep in mind

Newer and smaller than the other categories, and dependent on LiveBench's own harness.

Top 15
  1. 1DeepSeek V4.1 Flash (max)77.3%
  2. 2Claude Opus 5.5 (max)71.7%
  3. 3Claude Fable 5.1 (max)66.1%
  4. 4Claude Opus 5.5 (xhigh)65.3%
  5. 5Claude Opus 5 (max)65.2%
  6. 6Deepseek v4 Flash Vision65.1%
  7. 7Qwen3.8 Max (max)64.7%
  8. 8Muse Spark 1.3 (xhigh)64.1%
  9. 9Muse Wonder (max)62.7%
  10. 10Claude Fable 5 (max)62.2%
  11. 11Kimi K362.2%
  12. 12Qwen3.8-Flash-Next61.6%
  13. 13Qwen3.8 27B61.4%
  14. 14Smaug Flash61.1%
  15. 15GLM 5.360.9%
67 of 67
#
1DeepSeek V4.1 Flashmax77.3%61.22026-06-25
2Claude Opus 5.5max71.7%72.62026-06-25
3Claude Fable 5.1max66.1%70.52026-06-25
4Claude Opus 5.5xhigh65.3%70.02026-06-25
5Claude Opus 5max65.2%68.82026-06-25
6Deepseek v4 Flash Vision65.1%–2026-06-25
7Qwen3.8 Maxmax64.7%63.32026-06-25
8Muse Spark 1.3xhigh64.1%64.92026-06-25
9Muse Wondermax62.7%67.52026-06-25
10Claude Fable 5max62.2%69.32026-06-25
11Kimi K362.2%59.12026-06-25
12Qwen3.8-Flash-Next61.6%60.22026-06-25
13Qwen3.8 27B61.4%54.62026-06-25
14Smaug Flash61.1%–2026-06-25
15GLM 5.360.9%–2026-06-25
16Smaug Mini60.8%–2026-06-25
17Claude Sonnet 5xhigh59.4%59.02026-06-25
18Muse Spark 1.1xhigh58.5%61.62026-06-25
19Gemini 3.7 Flashhigh58.3%63.52026-06-25
20Muse Spark 1.2xhigh57.6%63.22026-06-25
21GPT-6 Astramax57.3%70.42026-06-25
22Claude Haiku 5.5max57.2%60.52026-06-25
23Mistral Large 4high57.2%49.82026-06-25
24Grok 4.657.0%57.82026-06-25
25GPT-6.1 Solxhigh56.8%67.72026-06-25
26GLM 5.3 Flash56.8%62.62026-06-25
27Grok 4.556.5%59.02026-06-25
28Claude Sonnet 5.5max56.3%69.32026-06-25
29GPT-5.6 Solmax56.2%68.02026-06-25
30GPT-5.6 Terramax55.0%64.42026-06-25
31DeepSeek V4 Pro55.0%53.52026-06-25
32GPT-6.1 Solmax54.5%69.92026-06-25
33Gemini 3.8 Flashhigh54.2%64.42026-06-25
34GPT-5.5xhigh54.0%66.12026-06-25
35Grok 4.7xhigh54.0%61.62026-06-25
36GPT-5.4xhigh53.8%64.72026-06-25
37GPT-6 Solmax52.9%66.42026-06-25
38Ox Alphamax52.6%–2026-06-25
39GLM 5.251.8%52.02026-06-25
40Claude Haiku 5.5xhigh51.4%61.52026-06-25
41GPT-6 Lunamax51.2%60.02026-06-25
42Claude Opus 4.7xhigh50.7%57.42026-06-25
43Claude Opus 4.8max50.5%63.32026-06-25
44GPT-5.2high50.3%56.02026-06-25
45GPT-5.2-Codex49.4%57.32026-06-25
46Gemini 3.5 Flashhigh49.0%63.12026-06-25
47Claude Opus 4.6high49.0%62.72026-06-25
48GPT-5.6 Lunamax48.4%59.92026-06-25
49Kimi K2.6thinking46.9%59.42026-06-25
50GPT-5.4 Nanoxhigh46.8%54.22026-06-25
51DeepSeek V4 Flash46.8%52.22026-06-25
52Grok Build 0.145.8%54.82026-06-25
53Kimi K2.7 Code45.7%55.82026-06-25
54Gemini 3.5 Flash Litehigh45.3%47.92026-06-25
55Gemini 3.1 Prohigh44.1%60.32026-06-25
56Qwen3.7max43.6%60.62026-06-25
57Gemini 3.6 Flashhigh43.4%62.02026-06-25
58Claude Sonnet 4.6medium42.6%–2026-06-25
59GPT-5.4 Minixhigh41.7%56.32026-06-25
60Qwen3.6 Plus41.4%58.82026-06-25

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LiveBench Agentic Coding leaderboard”, data as of 24 Oct 2026, https://www.benchleader.com/benchmarks/livebench_agentic_coding.

LiveBench Agentic Coding: questions

What does LiveBench Agentic Coding measure?
LiveBench's agentic coding category: the model works through multi-step coding tasks with tool use in an agent loop, on freshly written problems. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LiveBench Agentic Coding?
DeepSeek V4.1 Flash leads LiveBench Agentic Coding with 77.3% as of 24 Oct 2026, ahead of Claude Opus 5.5 at 71.7%.
How many models have LiveBench Agentic Coding results?
67 model configurations have a LiveBench Agentic Coding result on BenchLeader, all taken from LiveBench.
Who runs LiveBench Agentic Coding and how often is it updated?
LiveBench Agentic Coding is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LiveBench Agentic Coding count toward the BenchLeader Index?
No. LiveBench Agentic Coding is shown for reference but left out of the composite index.