BenchLeader

LMArena Hard Prompts leaderboard

As of 15 Nov 2026, the top score on LMArena Hard Prompts is 1547 by Gemini 4 Carbon (Google), ahead of Gemini 4 Argon at 1545, across 384 model configurations with a published result (Arena (LMArena), result dated 11 Nov 2026). The same pairwise votes as the text arena, restricted to prompts that LMArena's classifier judges hard: specific, complex, multi-step requests rather than casual chat. LMArena Hard Prompts counts toward the BenchLeader reasoning score.

Key facts (as of 15 Nov 2026)

  • Gemini 4 Carbon holds the top Arena Hard Prompts score at 1547 (Arena (LMArena), 11 Nov 2026).
  • Gemini 4 Argon is second on Arena Hard Prompts at 1545 (Arena (LMArena), 11 Nov 2026).
  • Arena Hard Prompts: Text arena rating on prompts judged hard (Arena (LMArena)).
  • 384 model configurations have a published Arena Hard Prompts result on BenchLeader (BenchLeader, 15 Nov 2026).
  • Arena Hard Prompts feeds the reasoning category of the BenchLeader Index (BenchLeader methodology).

Text arena rating on prompts judged hard.

Published by
Arena (LMArena)
Category
Reasoning
Index weight
0.5
Models
384
Data as of
15 Nov 2026

This week on Arena Hard Prompts

8 Nov 2026 to 15 Nov 2026, one row per model (its best setting on this board).

CC BY 4.0 — lmarena-ai/leaderboard-dataset on Hugging Face.

What the test looks like

The same pairwise votes as the text arena, restricted to prompts that LMArena's classifier judges hard: specific, complex, multi-step requests rather than casual chat.

How it is scored

Bradley-Terry rating on the hard-prompt subset, with style control.

What to keep in mind

Fewer votes per model than the overall board, so confidence intervals are wider. Hardness is judged by a classifier, not by outcome.

384 of 384
#
1Gemini 4 Carbonhigh154768.92026-11-11$4.00
2Gemini 4 Argonhigh154565.32026-11-11$4.00
3Muse Wondermax153160.82026-11-11$8.00
4Claude Opus 5.5high153066.02026-11-11$8.00
5Claude Opus 4.6high153062.52026-11-11$10.00
6Claude Fable 5high153063.92026-11-11$20.00
7Claude Opus 4.6152662.12026-11-11$10.00
8Claude Opus 4.7high152560.02026-11-11$10.00
9Claude Fable 5.1max152266.72026-11-11$20.00
10Claude Opus 4.7151963.12026-11-11$10.00
11Gemini 3.8 Flashhigh151859.42026-11-11$1.50
12Muse Spark 1.3max151760.62026-11-11$2.00
13Kimi K3max151660.22026-11-11$6.00
14Claude Opus 5high151564.42026-11-11$10.00
15NewGemini 4 Hydrogenhigh151463.42026-11-11$3.00
16Muse Spark 1.2xhigh151461.52026-11-11$2.00
17Claude Opus 5max151365.02026-11-11$10.00
18Claude Opus 4.8high151360.42026-11-11$10.00
19GPT-5.6 Solxhigh151163.92026-11-11$8.00
20Muse Spark 1.1151162.22026-11-11$2.00
21Gemini 3.7 Flashhigh151062.12026-11-11$1.50
22Gemini 3.1 Pro150859.92026-11-11$4.50
23GLM 5.3max150758.82026-11-11$2.15
24Qwen3.8 Maxmax150758.72026-11-11$3.00
25Muse Spark150664.22026-11-11–
26Claude Opus 4.8150558.72026-11-11$10.00
27GPT-6.1 Solmax150566.62026-11-11$4.00
28MiMo-V2.6-Pro150457.52026-11-11$0.548
29Kimi K3.1 Promax150459.92026-11-11$4.00
30Claude Sonnet 4.6150357.62026-11-11$6.00
31Gemini 3 Pro150260.02026-11-11–
32Gemini 3.6 Flashhigh150161.02026-11-11$1.50
33GPT-5.5high150163.52026-11-11$11.25
34Claude Opus 4.5high150159.72026-11-11$10.00
35GLM 5.3 Flash150057.82026-11-11$0.238
36GPT-6 Astramax150067.32026-11-11$20.00
37Claude Sonnet 5.5xhigh149962.52026-11-11$4.00
38GPT-5.5149961.02026-11-11$11.25
39DeepSeek V4.1 Flashmax149958.52026-11-11$0.525
40Claude Opus 4.5149957.12026-11-11$10.00
41GPT-5.2149756.72026-11-11$4.81
42GPT-5.4high149658.62026-11-11$5.63
43Gemini 3.5 Flashhigh149662.52026-11-11$3.38
44GLM 5.2max149658.92026-11-11$2.15
45Gemini 3.5 Flashmedium149561.82026-11-11$3.38
46Qwen3.7max149560.02026-11-11$3.75
47MiMo-V2.5-Pro149455.52026-11-11$1.50
48Gemini 3 Flash149356.52026-11-11$1.13
49Qwen3.5 Maxmax1492–2026-11-11–
50GPT-5.5 Instant149156.92026-11-11$11.25
51GPT-5.6 Terraxhigh149160.12026-11-11$4.50
52Claude Sonnet 5high149059.02026-11-11$4.00
53Grok 4.20thinking148958.32026-11-11$1.56
54GLM 5.1148955.42026-11-11$2.15
55Grok 4.5148859.32026-11-11$3.00
56ERNIE 5.11487–2026-11-11–
57Grok 4.20 11487–2026-11-11–
58GPT-5.4148757.92026-11-11$5.63
59Claude Sonnet 4.5high148757.92026-11-11$6.00
60Kimi K2.6148555.12026-11-11$1.71

AI reasoning and math benchmark leaderboard · All LLM benchmarks

Cite as: BenchLeader, “Arena Hard Prompts leaderboard”, data as of 15 Nov 2026, https://www.benchleader.com/benchmarks/lmarena_text_hard.

Arena Hard Prompts: questions

What does Arena Hard Prompts measure?
The same pairwise votes as the text arena, restricted to prompts that LMArena's classifier judges hard: specific, complex, multi-step requests rather than casual chat. Scores are reported in Elo-style ratings from pairwise comparisons; higher is better.
What is the top score on LMArena Hard Prompts?
As of 15 Nov 2026, the highest score on Arena Hard Prompts is 1547 by Gemini 4 Carbon (result dated 11 Nov 2026), ahead of Gemini 4 Argon at 1545. Scores come from Arena (LMArena) and are re-read daily.
Which AI model leads Arena Hard Prompts?
Gemini 4 Carbon leads Arena Hard Prompts with 1547 as of 15 Nov 2026 (result dated 11 Nov 2026), ahead of Gemini 4 Argon at 1545.
How many models have Arena Hard Prompts results?
384 model configurations have a Arena Hard Prompts result on BenchLeader, all taken from Arena (LMArena).
Who runs Arena Hard Prompts and how often is it updated?
Arena Hard Prompts is published by Arena (LMArena). BenchLeader re-reads the published results once a day and records the date each result was published.
Does Arena Hard Prompts count toward the BenchLeader Index?
Yes. Arena Hard Prompts contributes to the reasoning category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.