BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

AA-LCR

Long-context reasoning across ~100k-token document sets. Run by Artificial Analysis.

As of 25 Oct 2026, Kimi K3 leads AA-LCR on BenchLeader with 88.7%, ahead of Step 5 Preview at 88.3%, across 548 model configurations with a published result.

Published by
Artificial Analysis
Category
Long context
Index weight
1.0
Models
548
Data as of
25 Oct 2026
Newest result
23 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

What the test looks like

AA-LCR gives the model a set of documents totalling around 100,000 tokens and asks questions that require pulling together facts from several of them.

How it is scored

Percent of questions answered correctly, run by Artificial Analysis.

What to keep in mind

A single context length, so it does not show how performance degrades beyond it. Models with smaller windows are excluded or truncated.

Top 15
  1. 1Kimi K3 (max)88.7%
  2. 2Step 5 Preview88.3%
  3. 3Kimi K3.1 Pro86.3%
  4. 4MiMo-V2.6-Pro86.3%
  5. 5Claude Fable 5.1 (max)85.3%
  6. 6Claude Opus 5.5 (max)84.7%
  7. 7Claude Opus 5.5 (xhigh)84.7%
  8. 8Claude Fable 5.1 (medium)84.7%
  9. 9Claude Opus 5.5 (medium)84.3%
  10. 10GPT-5.5 (xhigh)84.3%
  11. 11GPT-5.5 (high)84.3%
  12. 12GPT-5.6 Sol (max)84.0%
  13. 13GPT-6.1 Sol (low)84.0%
  14. 14Gemini 3.8 Flash (medium)84.0%
  15. 15DeepSeek V4.1 Flash (max)84.0%
548 of 548
#
1Kimi K3max88.7%64.82026-07-16
2Step 5 Preview88.3%60.52026-09-18
3NewKimi K3.1 Pro86.3%66.92026-10-23
4MiMo-V2.6-Pro86.3%64.02026-09-21
5Claude Fable 5.1max85.3%70.52026-09-01
6Claude Opus 5.5max84.7%72.62026-09-22
7Claude Opus 5.5xhigh84.7%70.02026-09-22
8Claude Fable 5.1medium84.7%67.32026-09-01
9Claude Opus 5.5medium84.3%66.52026-09-22
10GPT-5.5xhigh84.3%66.12026-04-23
11GPT-5.5high84.3%65.32026-04-23
12GPT-5.6 Solmax84.0%68.02026-07-09
13GPT-6.1 Sollow84.0%64.92026-09-29
14Gemini 3.8 Flashmedium84.0%64.42026-09-02
15DeepSeek V4.1 Flashmax84.0%61.22026-09-10
16Muse Wonder Fullmax83.7%70.62026-10-12
17Claude Fable 5.1high83.7%69.42026-09-01
18GPT-6 Solmax83.7%66.42026-09-22
19GPT-6 Solhigh83.7%63.72026-09-22
20GPT-5.6 Lunamax83.7%59.92026-07-09
21GPT-6.1 Solmedium83.3%67.22026-09-29
22GPT-5.3-Codexxhigh83.3%63.82026-02-05
23GPT-6 Lunamax83.3%60.02026-09-22
24Muse Glimmerhigh83.3%51.42026-08-10
25Solar Mini 483.3%–2026-09-22
26GPT-6.1 Solmax83.0%69.92026-09-29
27Claude Fable 5.1xhigh83.0%69.52026-09-01
28Muse Spark 1.3max83.0%67.12026-09-02
29Muse Spark 1.3xhigh83.0%64.92026-09-02
30GPT-5.6 Terramax83.0%64.42026-07-09
31Gemini 3.7 Flashmedium83.0%63.12026-08-13
32GPT-5.5medium83.0%62.92026-04-23
33Agnes 2.5 Pro Beta83.0%59.12026-08-26
34MiniMax M383.0%56.42026-06-01
35Ling 3.1 Flash83.0%–2026-10-01
36Claude Sonnet 5.5max82.7%69.42026-09-28
37Claude Opus 5.5high82.7%69.42026-09-22
38DeepSeek V4.1 Flash 1017max82.7%64.02026-10-17
39GPT-5.2xhigh82.7%60.72025-12-11
40Claude Haiku 5.5max82.7%60.52026-10-07
41Claude Fable 5max82.3%69.32026-06-09
42GPT-6.1 Solhigh82.3%67.72026-09-29
43GPT-5.6 Solxhigh82.3%66.62026-07-09
44Claude Fable 5.1low82.3%65.02026-09-01
45GPT-6 Solmedium82.3%62.62026-09-22
46GPT-5.2-Codexxhigh82.3%59.42025-12-11
47GPT-5.4xhigh82.0%64.72026-03-05
48Claude Opus 5medium82.0%64.62026-07-24
49Gemini 3.1 Pro82.0%60.52026-02-19
50Claude Sonnet 5max82.0%58.42026-06-30
51Qwen3.8 27Bxhigh82.0%56.32026-08-14
52GPT-5.6 Solhigh81.7%66.32026-07-09
53Gemini 3.7 Flashhigh81.7%63.52026-08-13
54GPT-5.6 Lunaxhigh81.7%59.72026-07-09
55Nex-N2-Pro81.7%54.72026-06-02
56Gemini 3.8 Flashhigh81.3%64.42026-09-02
57GPT-6 Solxhigh81.3%64.32026-09-22
58Claude Opus 5low81.3%61.32026-07-24
59Deepseek v4 Flash Visionmax81.3%58.22026-08-21
60Mistral Large 481.3%56.72026-10-06

More long context benchmarks · All LLM benchmarks

Cite as: BenchLeader, “AA-LCR leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/aa_lcr.

AA-LCR: questions

What does AA-LCR measure?
AA-LCR gives the model a set of documents totalling around 100,000 tokens and asks questions that require pulling together facts from several of them. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads AA-LCR?
Kimi K3 leads AA-LCR with 88.7% as of 25 Oct 2026, ahead of Step 5 Preview at 88.3%.
How many models have AA-LCR results?
548 model configurations have a AA-LCR result on BenchLeader, all taken from Artificial Analysis.
Who runs AA-LCR and how often is it updated?
AA-LCR is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does AA-LCR count toward the BenchLeader Index?
Yes. AA-LCR contributes to the long context category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.