🚀Introducing Code Arena: the next generation of live coding evals for frontier AI models. Built to test how models plan, scaffold, debug, and build real web apps step-by-step. Try Claude, GPT-5, GLM-4.6 and Gemini in Code Arena today!

See how it performs in the Code Arena: the next generation of live coding evals for frontier AI models. Built to test how models plan, scaffold, debug, and build real web apps step-by-step. Head over to:

🚨MiniMax M2.1 by @MiniMax__AI is in the Code Arena! Test it out and evaluate it directly with other frontier AI. Don't forget to cast your votes in Battle Mode. We’ll publish the results soon. 🗳️

Test out Olmo-3.1-32B-Think vs. all the best Text AI models at: at

🚨New Model Update @Allen_AI’s Olmo-3.1-32B-Think is now available in the Text Arena! This open model is designed to perform strongly on reasoning, instruction following, and research-focused tasks. Bring your toughest prompts and see how it compares as community votes roll

Check out GPT-5.2 vs. all the other top frontier models at:

🚨BREAKING: Text Leaderboard Updates GPT-5.2 enters the Text leaderboard at #17, with a score of 1439. Compared to GPT-5.1, the model has improved by +2 points. It trails just one point behind GPT-5.2-high, which is optimized for expert-level reasoning and critical tasks.

Cast your vote and follow how GPT-5.2-Search and Grok-4.1-Fast-Search perform among the leading frontier AI search models:

🚨🌐Search Leaderboard Update @OpenAI’s GPT-5.2-Search and @xAI’s Grok-4.1-Fast-Search have landed on the Search Arena leaderboard. 🔹GPT-5.2-Search ranks #2 (score 1211) 🔹Grok-4.1-Fast-Search ranks #4 (score 1185) Both models debuted ahead of their predecessors, posting

Check out Reve V1.1 and its fast variant vs. all the top image AI models at:

🚨🖼️Image Edit Leaderboard Update Image AI progress isn’t slowing down. Reve V1.1 and its fast variant have entered the Image Edit leaderboard, both securing strong positions in the top 15. 🔸#8 Reve V1.1 (score of 1255) 🔸#15 Reve V1.1 fast (score of 1226) This represents a

Read more in our blog: https://t.co/Lro9tO76bj To install → pip install arena-rank

Arena-Rank is Apache-2.0, pip-installable, and open to contributions. Star and clone the repo: https://t.co/zsxFD0d8gg We’re excited to see what the community builds—and what we can learn together.

🔓📊We’re open-sourcing the code that powers the LMArena leaderboards. Today we’re releasing Arena-Rank, a Python package for paired-comparison ranking—so anyone can audit how our leaderboards are computed or use it to build their own. Arena-Rank implements: 🔹Bradley–Terry &

Every vote moves the leaderboard. Check out image AI models at:

“Create a square image of an open cookbook chapter, describing in detail how to make a Christmas turkey”

“Take this photo of Einstein and colourise it, add sunglasses and change his coat into an elaborate renaissance outfit”

“A portrait photo of Albert Einstein, he is wearing a Princeton University wool cardigan - On his shoulder is a luna moth - In his hand he is holding a single sunflower - Behind him there is a poster for the movie "Metropolis" and a Charizard poster - On a pedestal there is a

“Create a polaroid photograph of a historic image of Richard Feynman giving a physics lecture, with a hand written note on the photo”

“Create a candid 90s photo with a flash of teenagers at a house party”

“Create an image of flashy marketing campaign for decaf espresso aimed at sophisticated toddlers”