Dataset Viewer
Auto-converted to Parquet Duplicate
model_name
stringlengths
7
26
organization
stringlengths
3
9
license
stringclasses
6 values
score
float64
-0.15
0.14
score_ci_lower
float64
-0.16
0.12
score_ci_upper
float64
-0.14
0.15
observation_count
float64
98.4k
3.41M
session_count
float64
7.04k
62.3k
rank
int64
1
34
category
stringclasses
1 value
leaderboard_publish_date
stringdate
2026-07-13 00:00:00
2026-07-13 00:00:00
Claude Fable 5 (High)
anthropic
Proprietary
0.139412
0.12386
0.154964
456,112
16,059
1
overall
2026-07-13
GPT 5.6 Sol (xHigh)
openai
Proprietary
0.109384
0.071832
0.146935
476,633
7,881
2
overall
2026-07-13
Claude Opus 4.8 (Thinking)
anthropic
Proprietary
0.092791
0.079311
0.106271
1,387,896
33,392
3
overall
2026-07-13
GPT 5.5 (xHigh)
openai
Proprietary
0.082623
0.073895
0.091351
1,335,608
36,289
4
overall
2026-07-13
Claude Sonnet 5 (High)
anthropic
Proprietary
0.079961
0.062841
0.097082
1,954,629
23,640
5
overall
2026-07-13
Claude Opus 4.7 (Thinking)
anthropic
Proprietary
0.077339
0.065136
0.089542
1,123,188
34,357
6
overall
2026-07-13
Claude Opus 4.7
anthropic
Proprietary
0.076311
0.064092
0.088529
1,193,789
34,950
7
overall
2026-07-13
GPT 5.5 (High)
openai
Proprietary
0.071644
0.064143
0.079145
1,639,666
61,455
8
overall
2026-07-13
GLM 5.2 (Max)
zai
MIT
0.062383
0.051369
0.073397
1,182,660
31,993
9
overall
2026-07-13
GPT 5.5
openai
Proprietary
0.060482
0.053301
0.067663
1,285,496
62,335
10
overall
2026-07-13
Claude Opus 4.6
anthropic
Proprietary
0.058769
0.046475
0.071062
1,104,471
34,075
11
overall
2026-07-13
GPT 5.4 (High)
openai
Proprietary
0.058607
0.051179
0.066034
2,274,632
61,743
12
overall
2026-07-13
Grok 4.5
xai
Proprietary
0.04991
0.03123
0.06859
411,378
9,851
13
overall
2026-07-13
Claude Opus 4.8
anthropic
Proprietary
0.039042
0.023657
0.054426
1,220,337
31,463
14
overall
2026-07-13
Claude Sonnet 4.6
anthropic
Proprietary
0.019865
0.008632
0.031099
1,073,271
34,873
15
overall
2026-07-13
GLM 5.1
zai
MIT
0.011936
0.003751
0.020121
1,786,113
51,362
16
overall
2026-07-13
Qwen3.7 Plus
alibaba
Proprietary
-0.006084
-0.01991
0.007742
352,172
10,048
17
overall
2026-07-13
Gemini 3.1 Pro Preview
google
Proprietary
-0.007029
-0.013887
-0.000171
1,650,996
61,459
18
overall
2026-07-13
Kimi K2.7 Code
moonshot
Modified MIT
-0.007596
-0.027665
0.012473
218,714
7,468
19
overall
2026-07-13
Qwen3.7 Max
alibaba
Proprietary
-0.008098
-0.022022
0.005825
328,546
9,944
20
overall
2026-07-13
DeepSeek V4 Pro
deepseek
MIT
-0.011081
-0.024719
0.002557
395,529
10,335
21
overall
2026-07-13
Gemini 3.5 Flash (High)
google
Proprietary
-0.011193
-0.018899
-0.003487
1,629,463
45,991
22
overall
2026-07-13
Kimi K2.6
moonshot
Modified MIT
-0.020915
-0.041787
-0.000044
185,477
7,620
23
overall
2026-07-13
Mimo V2.5 Pro
xiaomi
MIT
-0.031678
-0.045716
-0.01764
338,915
10,379
24
overall
2026-07-13
Minimax M3
minimax
MiniMax Community License
-0.032297
-0.045866
-0.018728
527,131
9,913
25
overall
2026-07-13
DeepSeek V4 Flash
deepseek
MIT
-0.046382
-0.059407
-0.033356
436,228
10,051
26
overall
2026-07-13
Gemini 3.5 Flash (Medium)
google
Proprietary
-0.070385
-0.088328
-0.052442
223,036
7,041
27
overall
2026-07-13
Grok Build 0.1
xai
Proprietary
-0.082108
-0.089994
-0.074223
3,405,587
52,802
28
overall
2026-07-13
Grok 4.3 (High)
xai
Proprietary
-0.083119
-0.09114
-0.075098
1,033,143
41,665
29
overall
2026-07-13
Gemini 3 Flash
google
Proprietary
-0.085415
-0.093161
-0.077668
1,139,321
62,100
30
overall
2026-07-13
Minimax M2.7
minimax
Modified MIT
-0.114557
-0.130873
-0.098241
203,442
10,183
31
overall
2026-07-13
Gemma 4 31B
google
Apache 2.0
-0.135826
-0.150059
-0.121594
650,676
51,231
32
overall
2026-07-13
Nemotron 3 Ultra
nvidia
OpenMDW-1.1
-0.136807
-0.162464
-0.111149
98,417
9,511
33
overall
2026-07-13
Grok 4.3
xai
Proprietary
-0.153053
-0.163182
-0.142923
894,032
61,497
34
overall
2026-07-13

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
9,830