Dataset Viewer
Auto-converted to Parquet Duplicate
model_name
stringlengths
3
35
organization
stringlengths
3
9
license
stringclasses
8 values
score
float64
-0.24
0.16
score_ci_lower
float64
-0.28
0.13
score_ci_upper
float64
-0.2
0.19
observation_count
float64
169k
8.21M
session_count
float64
5.77k
97.9k
rank
int64
1
59
category
stringclasses
1 value
leaderboard_publish_date
stringdate
2026-09-05 00:00:00
2026-09-05 00:00:00
Claude Fable 5.1 (Max)
anthropic
Proprietary
0.158738
0.130363
0.187113
502,509
6,796
1
overall
2026-09-05
Claude Opus 5 (High)
anthropic
Proprietary
0.126833
0.109581
0.144086
2,700,251
22,594
2
overall
2026-09-05
Claude Opus 5 (Max)
anthropic
Proprietary
0.116686
0.097128
0.136245
2,231,613
18,112
3
overall
2026-09-05
Claude Fable 5 (High)
anthropic
Proprietary
0.102271
0.087664
0.116877
1,838,727
36,204
4
overall
2026-09-05
GPT 5.6 Sol (xHigh)
openai
Proprietary
0.093112
0.078231
0.107992
3,130,094
29,730
5
overall
2026-09-05
Claude Opus 4.8 (High)
anthropic
Proprietary
0.091671
0.076915
0.106426
1,908,535
38,148
6
overall
2026-09-05
Kimi K3 (Max)
moonshot
Kimi K3 license
0.07959
0.072779
0.086401
8,205,381
97,915
7
overall
2026-09-05
Claude Sonnet 5 (High)
anthropic
Proprietary
0.074105
0.05459
0.093619
2,855,637
28,662
8
overall
2026-09-05
GPT 5.5 (xHigh)
openai
Proprietary
0.07211
0.061311
0.08291
2,333,345
51,255
9
overall
2026-09-05
Hy4 preview
tencent
Apache 2.0
0.06917
0.054097
0.084243
1,235,248
14,405
10
overall
2026-09-05
Claude Opus 4.7 (High)
anthropic
Proprietary
0.06029
0.045374
0.075206
1,384,002
36,571
11
overall
2026-09-05
GLM 5.2 (Max)
zai
MIT
0.060273
0.052636
0.06791
3,954,149
69,385
12
overall
2026-09-05
Grok 4.5
xai
Proprietary
0.057276
0.045749
0.068804
1,732,012
35,237
13
overall
2026-09-05
Claude Opus 4.7
anthropic
Proprietary
0.056568
0.041465
0.07167
1,471,892
37,121
14
overall
2026-09-05
GPT 5.5 (High)
openai
Proprietary
0.055539
0.045041
0.066037
2,270,441
73,882
15
overall
2026-09-05
Qwen3.8 Max
alibaba
Proprietary
0.055081
0.044154
0.066009
1,570,439
20,691
16
overall
2026-09-05
Gemini 3.8 Flash (High)
google
Proprietary
0.054717
0.035497
0.073937
568,062
10,104
17
overall
2026-09-05
DeepSeek V4 Pro (High) (0813)
deepseek
MIT
0.054289
0.045276
0.063303
2,824,068
34,397
18
overall
2026-09-05
Grok 4.6 (xHigh)
xai
Proprietary
0.049767
0.037725
0.061809
1,888,481
18,119
19
overall
2026-09-05
Claude Opus 4.6
anthropic
Proprietary
0.046454
0.03232
0.060589
1,300,383
36,366
20
overall
2026-09-05
GPT 5.5
openai
Proprietary
0.043887
0.034887
0.052888
1,980,945
79,166
21
overall
2026-09-05
GLM 5.3 (Max)
zai
MIT
0.037962
0.030847
0.045077
4,229,280
53,898
22
overall
2026-09-05
GLM 5.3 Flash
zai
MIT
0.03674
0.026423
0.047057
1,883,310
18,992
23
overall
2026-09-05
Deepseek V4 Flash (High) (20260731)
deepseek
MIT
0.032923
0.024902
0.040943
5,639,654
57,183
24
overall
2026-09-05
GPT 5.6 Terra (xHigh)
openai
Proprietary
0.030433
0.018453
0.042412
1,048,556
17,972
25
overall
2026-09-05
GPT 5.4 (High)
openai
Proprietary
0.029942
0.020909
0.038975
3,506,012
78,359
26
overall
2026-09-05
Qwen3.8 Flash Next
alibaba
qwen-community-1.0
0.026073
0.013997
0.03815
2,852,587
20,296
27
overall
2026-09-05
GPT 5.6 Luna (xHigh)
openai
Proprietary
0.020882
0.009724
0.03204
1,563,830
19,435
28
overall
2026-09-05
Claude Opus 4.8
anthropic
Proprietary
0.018481
-0.00963
0.046592
1,438,529
33,690
29
overall
2026-09-05
Qwen 3.8 27B
alibaba
Apache 2.0
0.01027
0.000417
0.020122
2,449,390
22,585
30
overall
2026-09-05
Kimi K2.7 Code
moonshot
Modified MIT
0.007485
-0.018998
0.033969
409,905
11,142
31
overall
2026-09-05
DeepSeek V4 Pro
deepseek
MIT
0.005751
-0.005095
0.016597
1,570,167
32,155
32
overall
2026-09-05
Claude Sonnet 4.6
anthropic
Proprietary
0.004966
-0.007861
0.017793
1,445,308
39,399
33
overall
2026-09-05
Muse Spark 1.2 (xHigh)
meta
Proprietary
0.004615
-0.007206
0.016437
848,088
20,816
34
overall
2026-09-05
Gemini 3.7 Flash (High)
google
Proprietary
0.004334
-0.004908
0.013575
1,756,560
24,413
35
overall
2026-09-05
Kimi K2.6
moonshot
Modified MIT
-0.002944
-0.030911
0.025023
328,073
11,279
36
overall
2026-09-05
Muse Spark 1.1
meta
Proprietary
-0.014319
-0.020659
-0.00798
3,451,189
91,294
37
overall
2026-09-05
Qwen3.7 Max
alibaba
Proprietary
-0.016232
-0.025659
-0.006805
1,649,766
36,306
38
overall
2026-09-05
GLM 5.1
zai
MIT
-0.022229
-0.032116
-0.012343
3,299,854
71,952
39
overall
2026-09-05
Hy3
tencent
Apache 2.0
-0.024966
-0.038136
-0.011795
876,794
24,934
40
overall
2026-09-05
Gemini 3.5 Flash (High)
google
Proprietary
-0.034766
-0.043345
-0.026188
4,118,482
94,447
41
overall
2026-09-05
Gemini 3.6 Flash (High)
google
Proprietary
-0.035824
-0.047352
-0.024297
1,012,896
18,309
42
overall
2026-09-05
Gemini 3.1 Pro Preview
google
Proprietary
-0.037324
-0.046711
-0.027938
2,785,335
84,469
43
overall
2026-09-05
Minimax M3
minimax
MiniMax Community License
-0.037551
-0.046937
-0.028165
2,633,002
36,707
44
overall
2026-09-05
Mimo V2.5 Pro
xiaomi
MIT
-0.038124
-0.04794
-0.028308
1,678,566
37,593
45
overall
2026-09-05
Qwen3.7 Plus
alibaba
Proprietary
-0.038664
-0.051392
-0.025936
908,197
19,845
46
overall
2026-09-05
Gemini 3.5 Flash (Medium)
google
Proprietary
-0.047119
-0.062285
-0.031953
564,482
13,762
47
overall
2026-09-05
Inkling Small
thinky
Apache 2.0
-0.054274
-0.071533
-0.037015
369,725
11,299
48
overall
2026-09-05
Mistral Medium 3.5
mistral
Modified MIT
-0.071618
-0.087244
-0.055993
294,226
8,793
49
overall
2026-09-05
Inkling
thinky
Apache 2.0
-0.074742
-0.085441
-0.064043
1,546,849
40,752
50
overall
2026-09-05
Solar Pro 4
upstage
Proprietary
-0.09872
-0.134317
-0.063124
199,038
5,770
51
overall
2026-09-05
Grok 4.3 (High)
xai
Proprietary
-0.115708
-0.127438
-0.103979
1,646,025
62,791
52
overall
2026-09-05
Gemini 3.5 Flash Lite
google
Proprietary
-0.116381
-0.130307
-0.102456
670,210
23,215
53
overall
2026-09-05
Minimax M2.7
minimax
Modified MIT
-0.117077
-0.129012
-0.105142
842,531
37,001
54
overall
2026-09-05
Gemini 3 Flash
google
Proprietary
-0.121105
-0.132491
-0.109718
1,925,879
83,413
55
overall
2026-09-05
Grok Build 0.1
xai
Proprietary
-0.129476
-0.142973
-0.115978
5,586,049
74,459
56
overall
2026-09-05
Nemotron 3 Ultra
nvidia
OpenMDW-1.1
-0.14426
-0.171074
-0.117445
168,527
12,264
57
overall
2026-09-05
Grok 4.3
xai
Proprietary
-0.202638
-0.2215
-0.183776
1,299,696
82,746
58
overall
2026-09-05
Gemma 4 31B
google
Apache 2.0
-0.237006
-0.27549
-0.198523
703,573
56,661
59
overall
2026-09-05

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
27,958