Dataset Viewer
Auto-converted to Parquet Duplicate
model_name
stringlengths
3
35
organization
stringlengths
3
9
license
stringclasses
7 values
score
float64
-0.2
0.12
score_ci_lower
float64
-0.22
0.11
score_ci_upper
float64
-0.17
0.14
observation_count
float64
164k
6.72M
session_count
float64
5.2k
94.1k
rank
int64
1
50
category
stringclasses
1 value
leaderboard_publish_date
stringdate
2026-08-19 00:00:00
2026-08-19 00:00:00
Claude Opus 5 (High)
anthropic
Proprietary
0.124705
0.109296
0.140114
2,302,558
19,787
1
overall
2026-08-19
Claude Opus 5 (Max)
anthropic
Proprietary
0.120027
0.102005
0.138049
1,867,898
15,575
2
overall
2026-08-19
Claude Fable 5 (High)
anthropic
Proprietary
0.115663
0.098618
0.132707
1,376,967
31,667
3
overall
2026-08-19
Kimi K3 (Max)
moonshot
Kimi K3 license
0.104068
0.097883
0.110252
6,718,519
83,274
4
overall
2026-08-19
GPT 5.6 Sol (xHigh)
openai
Proprietary
0.097411
0.083472
0.11135
2,487,331
26,310
5
overall
2026-08-19
Claude Opus 4.8 (High)
anthropic
Proprietary
0.0955
0.080423
0.110577
1,632,487
35,201
6
overall
2026-08-19
GPT 5.5 (xHigh)
openai
Proprietary
0.085077
0.07577
0.094384
2,057,598
47,828
7
overall
2026-08-19
Claude Opus 4.7 (High)
anthropic
Proprietary
0.081196
0.068793
0.093599
1,341,740
36,165
8
overall
2026-08-19
GPT 5.5 (High)
openai
Proprietary
0.077399
0.06903
0.085767
2,231,149
73,102
9
overall
2026-08-19
Claude Opus 4.7
anthropic
Proprietary
0.074012
0.061557
0.086466
1,432,776
36,723
10
overall
2026-08-19
Claude Sonnet 5 (High)
anthropic
Proprietary
0.06617
0.044223
0.088118
2,495,203
25,827
11
overall
2026-08-19
Claude Opus 4.6
anthropic
Proprietary
0.06595
0.053974
0.077926
1,267,688
35,911
12
overall
2026-08-19
GPT 5.5
openai
Proprietary
0.06382
0.055861
0.07178
1,747,331
74,250
13
overall
2026-08-19
DeepSeek V4 Pro (High) (0813)
deepseek
MIT
0.06264
0.049883
0.075397
1,052,558
13,377
14
overall
2026-08-19
Qwen3.8 Max
alibaba
Proprietary
0.061998
0.048415
0.075581
780,252
11,800
15
overall
2026-08-19
Grok 4.5
xai
Proprietary
0.061714
0.050141
0.073287
1,463,803
30,936
16
overall
2026-08-19
GLM 5.2 (Max)
zai
MIT
0.058192
0.049542
0.066843
2,427,704
54,704
17
overall
2026-08-19
GPT 5.4 (High)
openai
Proprietary
0.049813
0.041808
0.057819
3,068,597
73,465
18
overall
2026-08-19
GPT 5.6 Luna (xHigh)
openai
Proprietary
0.040367
0.021643
0.059092
539,027
8,796
19
overall
2026-08-19
Deepseek V4 Flash (High) (20260731)
deepseek
MIT
0.039866
0.03201
0.047722
4,040,288
42,582
20
overall
2026-08-19
Gemini 3.7 Flash (High)
google
Proprietary
0.033215
0.023061
0.04337
1,233,390
18,109
21
overall
2026-08-19
GPT 5.6 Terra (xHigh)
openai
Proprietary
0.031916
0.019969
0.043864
788,069
14,592
22
overall
2026-08-19
Claude Sonnet 4.6
anthropic
Proprietary
0.028817
0.016936
0.040697
1,260,144
36,825
23
overall
2026-08-19
Claude Opus 4.8
anthropic
Proprietary
0.025069
0.00255
0.047588
1,404,239
33,237
24
overall
2026-08-19
Muse Spark 1.1
meta
Proprietary
0.009297
0.00319
0.015404
2,633,187
76,194
25
overall
2026-08-19
Kimi K2.7 Code
moonshot
Modified MIT
0.004341
-0.016367
0.025049
406,222
11,080
26
overall
2026-08-19
DeepSeek V4 Pro
deepseek
MIT
0.000533
-0.008239
0.009306
1,429,855
30,337
27
overall
2026-08-19
GLM 5.1
zai
MIT
-0.000049
-0.007502
0.007403
3,271,526
71,611
28
overall
2026-08-19
Gemini 3.5 Flash (High)
google
Proprietary
-0.006761
-0.01287
-0.000652
4,091,371
94,056
29
overall
2026-08-19
Qwen3.7 Max
alibaba
Proprietary
-0.007544
-0.016076
0.000987
1,341,595
31,904
30
overall
2026-08-19
Gemini 3.1 Pro Preview
google
Proprietary
-0.008806
-0.01565
-0.001962
2,610,514
81,574
31
overall
2026-08-19
Kimi K2.6
moonshot
Modified MIT
-0.011016
-0.033326
0.011295
325,483
11,226
32
overall
2026-08-19
Mimo V2.5 Pro
xiaomi
MIT
-0.02252
-0.031365
-0.013675
1,376,316
32,685
33
overall
2026-08-19
Hy3
tencent
Apache 2.0
-0.023825
-0.036241
-0.011409
675,512
19,741
34
overall
2026-08-19
Qwen3.7 Plus
alibaba
Proprietary
-0.025058
-0.037905
-0.01221
719,262
17,078
35
overall
2026-08-19
Gemini 3.6 Flash (High)
google
Proprietary
-0.030423
-0.04268
-0.018166
699,294
13,600
36
overall
2026-08-19
Minimax M3
minimax
MiniMax Community License
-0.031467
-0.039803
-0.023132
2,147,769
32,279
37
overall
2026-08-19
Gemini 3.5 Flash (Medium)
google
Proprietary
-0.038608
-0.052602
-0.024615
537,358
13,355
38
overall
2026-08-19
Inkling Small
thinky
Apache 2.0
-0.068235
-0.085773
-0.050697
262,109
8,513
39
overall
2026-08-19
Inkling
thinky
Apache 2.0
-0.0719
-0.081198
-0.062603
1,359,645
36,683
40
overall
2026-08-19
Mistral Medium 3.5
mistral
Modified MIT
-0.076021
-0.094785
-0.057257
178,503
6,018
41
overall
2026-08-19
Grok 4.3 (High)
xai
Proprietary
-0.090204
-0.098621
-0.081787
1,639,071
62,446
42
overall
2026-08-19
Gemini 3 Flash
google
Proprietary
-0.09132
-0.099477
-0.083163
1,913,089
83,040
43
overall
2026-08-19
Grok Build 0.1
xai
Proprietary
-0.094785
-0.103755
-0.085815
5,531,735
74,036
44
overall
2026-08-19
Solar Pro 4
upstage
Proprietary
-0.1082
-0.132076
-0.084324
180,321
5,198
45
overall
2026-08-19
Gemini 3.5 Flash Lite
google
Proprietary
-0.10822
-0.120505
-0.095935
566,628
20,524
46
overall
2026-08-19
Minimax M2.7
minimax
Modified MIT
-0.11865
-0.129587
-0.107713
702,439
32,465
47
overall
2026-08-19
Nemotron 3 Ultra
nvidia
OpenMDW-1.1
-0.153565
-0.176958
-0.130173
163,690
12,106
48
overall
2026-08-19
Grok 4.3
xai
Proprietary
-0.164526
-0.176756
-0.152296
1,299,377
82,414
49
overall
2026-08-19
Gemma 4 31B
google
Apache 2.0
-0.198161
-0.223841
-0.172482
708,361
56,608
50
overall
2026-08-19

Arena Leaderboard Dataset

Historical snapshots of the Arena leaderboard.

Usage

from datasets import load_dataset

# Load all historical text style control data
ds = load_dataset("lmarena-ai/leaderboard-dataset", "text_style_control", split="full")

# Load the current text style control leaderboard
ds = load_dataset("lmarena-ai/leaderboard-dataset", "text_style_control", split="latest")

# Filter to overall category
ds = load_dataset("lmarena-ai/leaderboard-dataset", "text_style_control", split="latest",
    filters=[("category", "==", "overall")]
)

# Track a specific model's rating over time
ds = load_dataset("lmarena-ai/leaderboard-dataset", "text_style_control", split="full",
    filters=[("category", "==", "overall"), ("model_name", "==", "gpt-4o-2024-05-13")],
    columns=["model_name", "rating", "rank", "leaderboard_publish_date"]
)

# Load raw (non-style-controlled) text ratings
ds = load_dataset("lmarena-ai/leaderboard-dataset", "text", split="full")

# Load the current Agent Arena leaderboard
ds = load_dataset("lmarena-ai/leaderboard-dataset", "agent", split="latest")

Subsets

Each arena is a separate subset, arenas with style control have an additional subset with a _style_control suffix .

  • text, text_style_control
  • vision, vision_style_control
  • search, search_style_control
  • document, document_style_control
  • webdev (Code Arena)
  • text_to_image
  • image_edit
  • text_to_video
  • image_to_video
  • video_edit
  • agent (Agent Arena aggregate), plus per-signal subsets: agent_bash_recovery_steps, agent_praise_complaint, agent_steerability, agent_task_outcome_explicit, agent_tool_hallucination

Splits

  • full: All historically published leaderboards
  • latest: Only the most recently published leaderboards

Notes

  • On January 9, 2024, the rating system was updated from Elo to Bradley-Terry.
  • On May 16, 2025, style control was made the default for text and vision arenas and the offset was adjusted to put the style control leaderboard on the same rating scale as non style control.
  • On July 23, 2025, frequency-based re-weighting was implemented.
  • Search and Webdev Arena data starts from their releases on the arena.ai (then lmarena.ai) on August 7 and November 12 2025 respectively.
  • Agent Arena data starts from its release on June 4, 2026.

For a complete record of all leaderboard methodology changes, see the Leaderboard Changelog.

Schema

Column Type Description
model_name string Model identifier
organization string Model creator/organization
license string Model license
rating float Arena Score
rating_lower float Lower confidence bound
rating_upper float Upper confidence bound
variance float Rating variance
vote_count int Number of battles for this model
rank int Rank within this leaderboard
category string Leaderboard category (e.g., overall, coding, math)
leaderboard_publish_date string Date that this score was published (YYYY-MM-DD)

Agent Arena subsets use IPS scores instead of Bradley-Terry:

Column Type Description
model_name string Model identifier
organization string Model creator/organization
license string Model license
score float IPS score (τ̂)
score_ci_lower float Lower 95% confidence bound
score_ci_upper float Upper 95% confidence bound
observation_count int Signal observations for this model
session_count int Distinct sessions (aggregate subset only)
rank int Rank within this leaderboard
category string Leaderboard category
leaderboard_publish_date string Date that this score was published (YYYY-MM-DD)
Downloads last month
36,426