8,094 comparisons among 28 agents.
One of 3 tables from the same source; this one is the t2i table. The others are listed under Related tables.
| Description | GenAI-Arena is TIGER-Lab's arena for generative vision models: a person writes a prompt, sees two anonymous models' outputs, and votes. The task an arena covers is encoded in the model name and nowhere else, so the deposit's votes are split into three tables (genai_arena_t2i, genai_arena_edit, genai_arena_t2v) and no vote crosses a task boundary. Image-generation votes: 8,094 votes between 28 named models, 2024-02-12 to 2025-01-07. agent_a wins 2,982, agent_b 2,968, 2,144 draws. No homefield column: slot assignment is balanced and the left side wins 50.07% of decided votes, z = +0.10. agent_a is the left model and agent_b the right, read from the record's `states` (the top-level `models` field is empty on every vote). `outcome` keeps the vote as recorded (leftvote, rightvote, tievote, bothbad_vote) where `winner` collapses the last two to draw. The source records carry an `ip` field, populated on about a quarter of rows; it is dropped. Those are private-range (10.x.x.x) load-balancer addresses, so they identify nobody -- and are useless as a rater key for the same reason, which is why these tables have no rater column. The prompts are dropped too. |
|---|---|
| Reference | TIGER-Lab. (2025). GenAI-Arena-logs [Data set]. Hugging Face. https://huggingface.co/datasets/TIGER-Lab/GenAI-Arena-logs |
| Licence | CC BY 4.0 |
| Source data | https://huggingface.co/datasets/TIGER-Lab/GenAI-Arena-logs |
| Comparisons | 8,094 |
|---|---|
| Agents | 28 |
agent_aagent_bdateoutcomescore_ascore_bwinner
Or load it directly in R (no account needed with irw 1.4.0 or later) or Python (needs a free Redivis account; the first call opens a browser to sign in):
# R
install.packages("irw")
library(irw)
df <- irw_fetch("genai_arena_t2i", source = "comp")
# Python
pip install irw
import irw
df = irw.fetch("genai_arena_t2i", source="comp")
Or in Stata, which reads the same no-account CSV as the button above:
* Stata
net install irw, from("https://raw.githubusercontent.com/itemresponsewarehouse/Stata-pkg/main/")
irw_fetch genai_arena_t2i, clear source(comp)
@dataset{tigerlab2025genaiarenalogs,
author = {TIGER-Lab},
title = {GenAI-Arena-logs},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/TIGER-Lab/GenAI-Arena-logs}
}
| IRW version | v712 |
|---|---|
| Redivis dataset | irw_competitions v7.1 |
| Redivis dataset DOI | 10.57761/3r1w-t660 |
| Manifest pin for this IRW version | v7.1 |
| Metadata source | irw_meta v68.0 |