Arena
A crowdsourced AI model arena where two models compete anonymously and users vote to determine the winner. Based on the results, public leaderboards for LLM, image, and code models are generated. It comes with built-in task templates such as landing pages, dashboards, and mini-games that can be directly tested in practice.
Updated: 2026-10-26
What is Arena?
Arena (arena.ai) is an AI model arena and leaderboard platform featuring "user voting to determine rankings". Its official website bills itself as the official AI ranking and LLM leaderboard. It places the power to judge "which model is better" in the hands of real users: the same prompt is sent to two anonymous models simultaneously, and you vote for the better response, with the voting results feeding into a public leaderboard shaped by the community. The platform covers three types of models: LLMs, image models and code models. It continuously updates its rankings through real-world evaluations and offers a variety of task templates, eliminating the need to write prompts from scratch for pre-selection testing.
Who is it for?
- Developers and product teams who need to select models but don't want to only rely on official benchmark scores; regular users who want to intuitively feel the differences in capabilities between various models; decision-makers who need to explain to their teams or clients "why choose this model". The voting threshold is low, making it friendly for non-technical users who are unfamiliar with benchmark testing terminology, but the website interaction is mainly in English, requiring a certain level of English reading ability.
Key features
5Anonymous Battle (Battle Mode)
Two models respond anonymously, with no brand displayed during voting. The results are more reflective of the quality of the responses themselves rather than the reputation of the vendors, avoiding preconceived biases.
Public Leaderboard
Community votes accumulate continuously to form separate rankings for LLM, image, and code models, which are updated driven by real-world evaluations rather than self-reported data from vendors.
Task Templates
Built-in common templates for creating landing pages, building dashboards, making playable mini-games, Design to Code (upload images for AI to generate them), building full-stack applications, setting up online stores, etc., to launch comparisons with one click.
File Input
Supports attaching files to conversations, bringing real materials into comparison scenarios, and making evaluations closer to actual usage.
Cutting-edge Experience
The platform's slogan is "Experience the frontier". You can directly run the latest models of various providers on the page to intuitively feel the differences in their capabilities.
How to use Arena
- The voting list reflects "what most people prefer" and is not entirely equivalent to objective ability.
- there may be biases in long-tail tasks.
- It is recommended to use the ranking as a filter: first run a round of prompt words close to your scenario using the built-in task templates, compare the response quality before narrowing down the candidate models.
- for models with insignificant differences in results, you can vote multiple times with the same prompt word to observe stability.
- The optimal model may vary for different task types (coding, copywriting, graphic design).
- It is recommended to conduct separate comparisons based on your high-frequency scenarios, and also keep an eye on dark horse models that have not been paid attention to in the list.
- For the final model selection conclusion, please conduct an additional round of verification combined with your actual tasks.
Things to know
Traffic & popularity
Data month Sep 2026
Arena receives about 22.3M visits per month and ranks #1,873 worldwide.
Monthly visits
12 monthsTraffic sources
- Direct68.4%
- Search19.1%
- Social6.4%
- AI assistants2.8%
- Referrals2.0%
- Other1.2%
Top countries
- India15.0%
- United States10.0%
- Russia9.4%
- Brazil4.3%
- Pakistan3.6%
Estimated from third-party traffic data.
Similar tools
imgsysimgsys is a generative image model arena from fal.ai that compares outputs from different models using the same prompt. It provides popularity rankings based on user votes with sample sizes indicated, covers both commercial and community models, and can be directly tried online.
SEAL LLM LeaderboardThe expert-driven evaluation list produced by the labs under Scale AI covers more than 20 benchmarks and over 100 models including agent programming, cutting-edge reasoning and safety alignment. It can be browsed by capability tiers and is continuously updated.
LLM StatsAn independent ranking that converts the intelligence level, speed and price of over 300 AI models into a single comprehensive score. The data is continuously updated from public benchmarks and real API metrics, supporting multi-task dimension ranking.
Design ArenaA crowdsourced AI design evaluation platform that sends the same request to multiple models for generation simultaneously, where users vote to determine the winners and compile category rankings, covering websites, slides, images, videos, games, and 3D assets.
Klu.aiAn automatic evaluation platform for cutting-edge models targeting business scenarios. The official website slogan is "Find your cutting edge". Generate evaluation datasets according to the workflow you describe, track model updates, and rerun comparisons of quality and cost each time a new model is released.
PromptPilotByteDance's Volcano Engine has launched a prompt engineering platform, which, according to the official, covers prompt generation, debugging and batch evaluation. It supports scenarios such as text understanding, visual understanding and multi-turn dialogue, and allows uploading task data to automatically optimize prompts and call them through APIs.