BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases15

LiveBench 2.0

Average of LiveBench 2.0's seven suite scores (long agentic, reasoning, coding, browser use, data analysis, research, design), on the current release.

As of 21 Oct 2026, Claude Fable 5.1 leads LiveBench 2.0 on BenchLeader with 88.0%, ahead of Claude Opus 5.5 at 86.5%, across 48 model configurations with a published result.

Published by
LiveBench
Category
Composite
Index weight
1.0
Models
48
Data as of
21 Oct 2026

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

Not comparable with the earlier test of the same name, whose final scores stay on LiveBench 1.0.

What the test looks like

LiveBench is a contamination-focused benchmark refreshed by release. LiveBench 2.0, released on 14 October 2026, has seven suites: five agentic ones (long agentic work, coding, browser use, research, design), where the model works through real tasks with tools, and two short question sets (reasoning, data analysis).

How it is scored

The average of LiveBench 2.0's suite scores on the current release, each a mean task score from 0 to 100 from a single run, graded automatically.

What to keep in mind

A different test from LiveBench 1.0, so 2.0 scores start on 19 October 2026 and are not comparable with 1.0's (see LiveBench 1.0). Within 2.0 the question set can still change between releases; BenchLeader records the date of each result.

Top 15
  1. 1Claude Fable 5.188.0%
  2. 2Claude Opus 5.586.5%
  3. 3GPT-6 Astra86.1%
  4. 4GPT-6.1 Sol85.4%
  5. 5Gemini 4 Argon84.1%
  6. 6Claude Fable 583.6%
  7. 7Claude Sonnet 5.582.5%
  8. 8Claude Opus 582.1%
  9. 9GPT-6 Sol81.9%
  10. 10DeepSeek V4.1 Flash 101779.2%
  11. 11Muse Wonder78.1%
  12. 12DeepSeek V4.1 Flash78.0%
  13. 13Claude Opus 4.876.8%
  14. 14GPT-5.6 Sol76.3%
  15. 15Claude Haiku 5.575.6%
48 of 48
#
1Claude Fable 5.188.0%66.22026-10-14
2Claude Opus 5.586.5%61.92026-10-14
3GPT-6 Astra86.1%67.92026-10-14
4GPT-6.1 Sol85.4%63.32026-10-14
5Gemini 4 Argon84.1%63.32026-10-14
6Claude Fable 583.6%67.12026-10-14
7Claude Sonnet 5.582.5%–2026-10-14
8Claude Opus 582.1%65.42026-10-14
9GPT-6 Sol81.9%–2026-10-14
10NewDeepSeek V4.1 Flash 101779.2%–2026-10-14
11Muse Wonder78.1%63.62026-10-14
12DeepSeek V4.1 Flash78.0%–2026-10-14
13Claude Opus 4.876.8%60.12026-10-14
14GPT-5.6 Sol76.3%55.92026-10-14
15Claude Haiku 5.575.6%–2026-10-14
16GPT-5.6 Terra75.0%52.62026-10-14
17Grok 4.674.9%57.82026-10-14
18MiMo-V2.6-Pro74.9%64.12026-10-14
19Muse Spark 1.374.8%58.32026-10-14
20Grok 4.774.1%–2026-10-14
21NewMiniMax-M3.1-Flash73.1%61.92026-10-14
22Claude Sonnet 573.0%56.02026-10-14
23GLM 5.3 Flash70.6%62.72026-10-14
24Grok 4.570.1%58.92026-10-14
25GPT-6 Luna69.4%–2026-10-14
26Qwen3.8 Flash68.8%–2026-10-14
27GPT-5.568.7%60.22026-10-14
28Gemini 3.8 Flash67.3%55.62026-10-14
29Gemini 3.7 Flash67.2%53.22026-10-14
30GPT-5.6 Luna66.1%47.92026-10-14
31Claude Sonnet 4.665.1%57.12026-10-14
32Qwen3.8 Maxmax64.2%63.32026-10-14
33Deepseek v4 Flash Vision63.6%–2026-10-14
34Kimi K363.4%59.12026-10-14
35Qwen3.8 27B62.7%54.52026-10-14
36Muse Spark 1.262.4%–2026-10-14
37Step 5 Preview62.0%60.62026-10-14
38GLM 5.360.7%–2026-10-14
39NewMuse Spark Open57.6%–2026-10-14
40NewBeam 1.055.5%54.62026-10-14
41Gemini 3.1 Pro54.9%60.72026-10-14
42MiniMax M353.7%56.52026-10-14
43Nemotron Super VL 3.551.5%–2026-10-14
44Mistral Large 447.7%56.72026-10-14
45Claude Haiku 4.546.0%45.32026-10-14
46Gemma 4 31B41.5%48.32026-10-14
47Nemotron Ultra 340.8%–2026-10-14
48Nemotron Lightning 3.528.7%–2026-10-14

More composite benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LiveBench 2.0 leaderboard”, https://www.benchleader.com/benchmarks/livebench2_global, data as of 21 Oct 2026.

LiveBench 2.0: questions

What does LiveBench 2.0 measure?
LiveBench is a contamination-focused benchmark refreshed by release. LiveBench 2.0, released on 14 October 2026, has seven suites: five agentic ones (long agentic work, coding, browser use, research, design), where the model works through real tasks with tools, and two short question sets (reasoning, data analysis). Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LiveBench 2.0?
Claude Fable 5.1 leads LiveBench 2.0 with 88.0% as of 21 Oct 2026, ahead of Claude Opus 5.5 at 86.5%.
How many models have LiveBench 2.0 results?
48 model configurations have a LiveBench 2.0 result on BenchLeader, all taken from LiveBench.
Who runs LiveBench 2.0 and how often is it updated?
LiveBench 2.0 is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LiveBench 2.0 count toward the BenchLeader Index?
Yes. LiveBench 2.0 contributes to the composite category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.