BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

LiveBench Language

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

As of 25 Oct 2026, Claude Fable 5 leads LiveBench Language on BenchLeader with 90.7%, ahead of GPT-6.1 Sol at 90.1%, across 67 model configurations with a published result.

Published by
LiveBench
Category
Instruction following
Index weight
Reference only
Models
67
Data as of
25 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 language tasks: word puzzles, typo correction and plot unscrambling built from recent text.

How it is scored

Percent correct, 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

Small and idiosyncratic; a good check that a model handles text carefully rather than a measure of writing quality.

Top 15
  1. 1Claude Fable 5 (max)90.7%
  2. 2GPT-6.1 Sol (max)90.1%
  3. 3Claude Fable 5.1 (max)89.5%
  4. 4GPT-6 Astra (max)89.4%
  5. 5Claude Opus 5 (max)88.7%
  6. 6GPT-6.1 Sol (xhigh)88.7%
  7. 7Gemini 3.8 Flash (high)87.8%
  8. 8GPT-5.6 Sol (max)87.7%
  9. 9GPT-5.5 (xhigh)87.4%
  10. 10Claude Opus 5.5 (max)86.3%
  11. 11Kimi K385.5%
  12. 12Claude Opus 5.5 (xhigh)85.5%
  13. 13Gemini 3.7 Flash (high)85.5%
  14. 14Gemini 3.1 Pro (high)85.4%
  15. 15GPT-6 Sol (max)85.3%
67 of 67
#
1Claude Fable 5max90.7%69.32026-06-25
2GPT-6.1 Solmax90.1%69.92026-06-25
3Claude Fable 5.1max89.5%70.52026-06-25
4GPT-6 Astramax89.4%70.42026-06-25
5Claude Opus 5max88.7%68.82026-06-25
6GPT-6.1 Solxhigh88.7%67.72026-06-25
7Gemini 3.8 Flashhigh87.8%64.42026-06-25
8GPT-5.6 Solmax87.7%68.02026-06-25
9GPT-5.5xhigh87.4%66.12026-06-25
10Claude Opus 5.5max86.3%72.62026-06-25
11Kimi K385.5%58.82026-06-25
12Claude Opus 5.5xhigh85.5%70.02026-06-25
13Gemini 3.7 Flashhigh85.5%63.52026-06-25
14Gemini 3.1 Prohigh85.4%60.32026-06-25
15GPT-6 Solmax85.3%66.42026-06-25
16Gemini 3.5 Flashhigh84.6%63.12026-06-25
17Gemini 3.6 Flashhigh83.9%62.02026-06-25
18Grok 4.683.7%57.82026-06-25
19Claude Sonnet 5.5xhigh83.4%67.92026-06-25
20Claude Opus 4.6high83.3%62.72026-06-25
21Muse Wondermax82.9%67.52026-06-25
22GPT-5.6 Terramax82.9%64.42026-06-25
23Grok 4.582.8%59.02026-06-25
24Muse Spark 1.3xhigh82.8%64.92026-06-25
25GPT-5.4xhigh82.6%64.72026-06-25
26DeepSeek V4 Pro82.1%53.42026-06-25
27Claude Opus 4.5high81.3%59.22026-06-25
28DeepSeek V4.1 Flashmax81.2%61.22026-06-25
29Deepseek v4 Flash Vision80.4%–2026-06-25
30Grok 4.7xhigh80.1%61.62026-06-25
31GLM 5.379.9%–2026-06-25
32GPT-5.2high79.8%56.02026-06-25
33Qwen3.7max79.7%60.62026-06-25
34Qwen3.8 Maxmax79.7%63.32026-06-25
35Claude Opus 4.8max79.7%63.32026-06-25
36Smaug Flash79.2%–2026-06-25
37DeepSeek V4 Flash79.2%52.22026-06-25
38Muse Spark 1.2xhigh78.6%63.22026-06-25
39Claude Sonnet 5.5max78.0%69.42026-06-25
40Claude Opus 4.7xhigh77.9%57.42026-06-25
41Kimi K2.7 Code77.9%55.82026-06-25
42GLM 5.3 Flash77.3%62.72026-06-25
43Smaug Mini77.0%–2026-06-25
44MiniMax M376.8%56.42026-06-25
45GLM 5.276.2%51.82026-06-25
46Claude Sonnet 4.6medium76.1%–2026-06-25
47Kimi K2.6thinking75.1%59.42026-06-25
48Qwen3.6 Plus75.0%58.82026-06-25
49Claude Sonnet 5xhigh75.0%59.02026-06-25
50Qwen3.8-Flash-Next74.6%60.32026-06-25
51Qwen3.8 27B74.3%54.62026-06-25
52Muse Spark 1.1xhigh74.3%61.62026-06-25
53GPT-6 Lunamax73.8%60.02026-06-25
54GPT-5.2-Codex73.7%57.32026-06-25
55Grok 4.373.6%54.62026-06-25
56GPT-5.6 Lunamax72.6%59.92026-06-25
57Grok Build 0.172.5%54.82026-06-25
58Gemini 3.5 Flash Litehigh71.8%47.92026-06-25
59GPT-5.4 Minixhigh71.0%56.32026-06-25
60Nemotron 3 Ultra 550B A55B70.8%55.52026-06-25

More instruction following benchmarks · All LLM benchmarks

Cite as: BenchLeader, “LiveBench Language leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/livebench_language.

LiveBench Language: questions

What does LiveBench Language measure?
LiveBench's language tasks: word puzzles, typo correction and plot unscrambling built from recent text. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads LiveBench Language?
Claude Fable 5 leads LiveBench Language with 90.7% as of 25 Oct 2026, ahead of GPT-6.1 Sol at 90.1%.
How many models have LiveBench Language results?
67 model configurations have a LiveBench Language result on BenchLeader, all taken from LiveBench.
Who runs LiveBench Language and how often is it updated?
LiveBench Language is published by LiveBench. BenchLeader re-reads the published results once a day and records the date each result was published.
Does LiveBench Language count toward the BenchLeader Index?
No. LiveBench Language is shown for reference but left out of the composite index.