Dataset Viewer
Auto-converted to Parquet Duplicate
model_name
stringlengths
7
26
organization
stringclasses
10 values
license
stringclasses
6 values
score
float64
-0.16
0.14
score_ci_lower
float64
-0.17
0.13
score_ci_upper
float64
-0.15
0.16
observation_count
float64
79.9k
3.02M
session_count
float64
5.24k
57.4k
rank
int64
1
29
category
stringclasses
1 value
leaderboard_publish_date
stringdate
2026-07-07 00:00:00
2026-07-07 00:00:00
Claude Fable 5 (High)
anthropic
Proprietary
0.140627
0.125016
0.156237
458,601
16,059
1
overall
2026-07-07
Claude Opus 4.8 (Thinking)
anthropic
Proprietary
0.097817
0.08477
0.110864
1,301,969
32,173
2
overall
2026-07-07
GPT 5.5 (xHigh)
openai
Proprietary
0.085161
0.075975
0.094346
1,135,901
31,540
3
overall
2026-07-07
Claude Opus 4.7
anthropic
Proprietary
0.083072
0.070768
0.095376
1,096,881
33,734
4
overall
2026-07-07
Claude Opus 4.7 (Thinking)
anthropic
Proprietary
0.082686
0.070307
0.095065
1,038,862
33,061
5
overall
2026-07-07
Claude Sonnet 5 (High)
anthropic
Proprietary
0.077152
0.064824
0.08948
1,646,453
22,511
6
overall
2026-07-07
GPT 5.5 (High)
openai
Proprietary
0.073474
0.06596
0.080988
1,466,428
56,780
7
overall
2026-07-07
GPT 5.5
openai
Proprietary
0.06736
0.059979
0.07474
1,144,443
57,373
8
overall
2026-07-07
GLM 5.2 (Max)
zai
MIT
0.066224
0.054233
0.078214
1,024,052
28,435
9
overall
2026-07-07
GPT 5.4 (High)
openai
Proprietary
0.06584
0.05811
0.07357
2,029,638
56,860
10
overall
2026-07-07
Claude Opus 4.6
anthropic
Proprietary
0.062421
0.050478
0.074365
1,039,695
32,872
11
overall
2026-07-07
Claude Opus 4.8
anthropic
Proprietary
0.046395
0.0308
0.061989
1,143,550
30,252
12
overall
2026-07-07
Claude Sonnet 4.6
anthropic
Proprietary
0.024761
0.013897
0.035625
1,004,646
33,639
13
overall
2026-07-07
GLM 5.1
zai
MIT
0.015842
0.007112
0.024573
1,503,694
46,604
14
overall
2026-07-07
Gemini 3.1 Pro Preview
google
Proprietary
-0.004125
-0.011042
0.002792
1,480,477
56,839
15
overall
2026-07-07
Kimi K2.7 Code
moonshot
Modified MIT
-0.010588
-0.024114
0.002937
909,760
28,613
16
overall
2026-07-07
DeepSeek V4 Flash
deepseek
MIT
-0.011888
-0.024717
0.000941
770,344
40,268
17
overall
2026-07-07
Kimi K2.6
moonshot
Modified MIT
-0.01897
-0.028272
-0.009667
1,072,910
48,744
18
overall
2026-07-07
Minimax M3
minimax
MiniMax Community License
-0.022575
-0.033687
-0.011463
1,486,147
34,335
19
overall
2026-07-07
DeepSeek V4 Pro
deepseek
MIT
-0.027749
-0.040497
-0.015001
876,083
30,757
20
overall
2026-07-07
Qwen 3.6 Plus
alibaba
Proprietary
-0.047278
-0.057858
-0.036699
821,882
43,467
21
overall
2026-07-07
Grok 4.3 (High)
xai
Proprietary
-0.076431
-0.084444
-0.068417
902,337
36,897
22
overall
2026-07-07
Grok Build 0.1
xai
Proprietary
-0.078144
-0.086224
-0.070064
3,015,968
47,957
23
overall
2026-07-07
gemini-3.5-flash
google
Proprietary
-0.079072
-0.100756
-0.057389
159,281
5,235
24
overall
2026-07-07
Minimax M2.7
minimax
Modified MIT
-0.081799
-0.09009
-0.073507
1,059,789
46,591
25
overall
2026-07-07
Gemini 3 Flash
google
Proprietary
-0.083277
-0.091288
-0.075265
1,003,002
57,173
26
overall
2026-07-07
Nemotron 3 Ultra
nvidia
OpenMDW-1.1
-0.117467
-0.146222
-0.088712
79,852
8,556
27
overall
2026-07-07
Gemma 4 31B
google
Apache 2.0
-0.130651
-0.145538
-0.115764
564,560
46,561
28
overall
2026-07-07
Grok 4.3
xai
Proprietary
-0.156904
-0.167101
-0.146707
808,646
56,628
29
overall
2026-07-07

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
9,406

Space using lmarena-ai/leaderboard-dataset 1