Celebrating Community Impact: 3M+ votes, 400+ models, and 300+ pre-release tests

To date, the community has evaluated over 400+ public models on LMArena as well as 300+ pre-release tests. Tens of millions of battle pairings have been served to users across the world, and each vote has shaped real-world AI performance and development. Around this time two years ago, the community helped us publish our very first leaderboard rankings:

The first Chatbot Arena Leaderboard (May 2023)

It’s amazing to see how far we’ve come together since then! Today we’re celebrating some fun figures outlining the scale of this community evaluation effort, showcasing how it has helped LLMs develop:

Current leaderboard (as of April 2025):

Model Releases on LMArena

Our community wants to explore and test the frontier of AI. Guided by their interest, we track and release the best models as they come. Big, well-known labs like Google, Meta, Alibaba and OpenAI are represented, as they are incredibly active in releasing models the community is eager to test. At the same time, thanks to community support and requests, open-source and emerging labs have a platform here too. LMArena exists to serve the community’s curiosity and ability to explore all types of models.

As part of our commitment to AI transparency, we’ve also released public datasets, and shared evaluation methodology and published papers to help everyone in the ecosystem learn and build better AI. You can explore and learn alongside the community by cross-checking the model release numbers above. We also share 20% of this data back with model providers to help them improve their models from our community’s perspective. Check out more of our ongoing research on our blog.

Sampling the Model Pairs

From the start, the LMArena experience has been community-driven. One trend that continues to come through in the feedback loud and clear is that a valuable experience involves having frequent access to the best models, and being a part of the first to access the newest ones. To meet the community expectations, our sampling policy focuses on those two areas:

  1. Preserve the community experience by serving the best models.

  2. Make it easy to converge on new models as they arrive.

To achieve this, we upsample the best models, and new models, so they get seen more frequently in battles and provide the most value back to the community. The way we prioritize models is just one example of how community input shapes every aspect of the platform.

Supporting Open-Source with Open Data

In addition to having ungated access to top models, the community remains especially excited and always passionate about testing the latest and greatest open-source models. That said, the leaderboard is the place where the best models are revealed. Every model steps into the arena, but only the best rise to the top. Open source or proprietary, it doesn’t matter, the community calls the winners.

We proudly support open-source models with nearly 41% of battles involving an open model. Our community not only champions open-source initiatives but also understands firsthand how much more cost efficient and customizable these models can be for real-world applications. Moreover, we have been releasing open data since Day 1, including over 1.5 millions prompts and 200K feedback votes to support open research. Expect more coming soon!

Data release on our HuggingFace

Join the Community

We’re going to keep working to make LMArena an even more inclusive and open experience. If you’ve been following our journey, you know we’re actively testing a new experience in Beta to be more accessible to everyone. We’re also going to continue to expand the range of models we test, improving ways for new teams to participate, and building features to make the evaluation experience better for the whole ecosystem. If you’re building something new or curious to explore the frontier of AI with others, we’d love for you to help shape the future of AI with us.