BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases15

LiveBench Long Agentic

LiveBench 2.0 long-horizon agentic suite.

As of 21 Oct 2026, Claude Opus 5.5 leads LiveBench Long Agentic on BenchLeader with 85.0%, ahead of GPT-6 Astra at 82.4%, across 48 model configurations with a published result.

Published by
LiveBench
Category
Agents & tools
Index weight
Reference only
Models
48
Data as of
21 Oct 2026

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

What the test looks like

LiveBench's long agentic suite: multi-step work where early decisions have to hold up through later turns, months or test cases.

How it is scored

Mean task score for the suite, 0–100, from a single run.

What to keep in mind

Long-horizon and harness-dependent; a zero can mean the run failed rather than a partial attempt.

Top 15
  1. 1Claude Opus 5.585.0%
  2. 2GPT-6 Astra82.4%
  3. 3Claude Fable 5.180.8%
  4. 4Claude Sonnet 5.578.8%
  5. 5GPT-6.1 Sol73.0%
  6. 6Gemini 4 Argon69.9%
  7. 7GPT-6 Sol66.9%
  8. 8Claude Opus 562.1%
  9. 9Claude Haiku 5.557.6%
  10. 10Claude Fable 553.2%
  11. 11Muse Wonder53.2%
  12. 12DeepSeek V4.1 Flash 101750.9%
  13. 13Grok 4.750.1%
  14. 14GPT-5.6 Sol48.5%
  15. 15GPT-5.6 Terra48.4%
48 of 48
#
1Claude Opus 5.585.0%61.92026-10-14
2GPT-6 Astra82.4%67.92026-10-14
3Claude Fable 5.180.8%66.22026-10-14
4Claude Sonnet 5.578.8%–2026-10-14
5GPT-6.1 Sol73.0%63.32026-10-14
6Gemini 4 Argon69.9%63.32026-10-14
7GPT-6 Sol66.9%–2026-10-14
8Claude Opus 562.1%65.42026-10-14
9Claude Haiku 5.557.6%–2026-10-14
10Claude Fable 553.2%67.12026-10-14
11Muse Wonder53.2%63.62026-10-14
12NewDeepSeek V4.1 Flash 101750.9%–2026-10-14
13Grok 4.750.1%–2026-10-14
14GPT-5.6 Sol48.5%55.92026-10-14
15GPT-5.6 Terra48.4%52.62026-10-14
16Claude Opus 4.848.3%60.12026-10-14
17GPT-6 Luna48.1%–2026-10-14
18MiMo-V2.6-Pro47.5%64.12026-10-14
19GPT-5.544.6%60.22026-10-14
20Kimi K342.8%59.12026-10-14
21Qwen3.8 Maxmax42.3%63.32026-10-14
22Grok 4.640.7%57.82026-10-14
23Muse Spark 1.340.4%58.32026-10-14
24DeepSeek V4.1 Flash40.1%–2026-10-14
25Claude Sonnet 539.0%56.02026-10-14
26Deepseek v4 Flash Vision39.0%–2026-10-14
27Qwen3.8 Flash37.1%–2026-10-14
28GLM 5.335.9%–2026-10-14
29NewMiniMax-M3.1-Flash34.7%61.92026-10-14
30GPT-5.6 Luna30.4%47.92026-10-14
31Gemini 3.8 Flash30.1%55.62026-10-14
32GLM 5.3 Flash28.4%62.72026-10-14
33Gemini 3.7 Flash28.3%53.22026-10-14
34Grok 4.527.9%58.92026-10-14
35Qwen3.8 27B27.9%54.52026-10-14
36Step 5 Preview27.5%60.62026-10-14
37Gemini 3.1 Pro23.9%60.72026-10-14
38Claude Sonnet 4.623.1%57.12026-10-14
39Muse Spark 1.221.3%–2026-10-14
40NewMuse Spark Open18.9%–2026-10-14
41NewBeam 1.017.3%54.62026-10-14
42Nemotron Super VL 3.515.8%–2026-10-14
43Gemma 4 31B13.3%48.32026-10-14
44MiniMax M312.4%56.52026-10-14
45Mistral Large 411.8%56.72026-10-14
46Nemotron Ultra 311.2%–2026-10-14
47Claude Haiku 4.511.1%45.32026-10-14
48Nemotron Lightning 3.58.3%–2026-10-14

More agents & tools benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LiveBench Long Agentic leaderboard”, https://www.benchleader.com/benchmarks/livebench_long_agentic, data as of 21 Oct 2026.

LiveBench Long Agentic: questions

What does LiveBench Long Agentic measure?
LiveBench's long agentic suite: multi-step work where early decisions have to hold up through later turns, months or test cases. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LiveBench Long Agentic?
Claude Opus 5.5 leads LiveBench Long Agentic with 85.0% as of 21 Oct 2026, ahead of GPT-6 Astra at 82.4%.
How many models have LiveBench Long Agentic results?
48 model configurations have a LiveBench Long Agentic result on BenchLeader, all taken from LiveBench.
Who runs LiveBench Long Agentic and how often is it updated?
LiveBench Long Agentic is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LiveBench Long Agentic count toward the BenchLeader Index?
No. LiveBench Long Agentic is shown for reference but left out of the composite index.