Code Arena

“this site has changed my life im autistic and permentaly sick with ibs and have little to no income thanks to arena.ai agent iv startet å buiness and now im getting alot done thanks so much from the bottom of my heart i hope its free in the future too”

“Previously they placed GPT 5.5 below Meta's Muse Spark in terms of coding ability. This latest benchmark they've released with Grok Imagine surpassing Seedance video generation... if anyone is currently using both it's fair to say this is objectively dishonest.”

“Insane that it beats chatGPT and Gemini by such a landslide”

“Public benchmark sucks. Check GPT-5.5 on Arc AGI 2, It's 84% while Kimi is 11%. https://arcprize.org/leaderboard”

“this site has changed my life im autistic and permentaly sick with ibs and have little to no income thanks to arena.ai agent iv startet å buiness and now im getting alot done thanks so much from the bottom of my heart i hope its free in the future too”

“Anyone else actually run this on real work or just vibes so far? I'm working with it for work since last week (some good test refactoring and it's decent). I never really used opus much (only sonnet ) so hard to compare. I did the same work with sonnet. It's faster but ran out of quota after 3 "classes" (while glm is muuuch more generous)”

“Fantastyczna strona do sprawdzania najnowszych modeli za darmo. Testowania. Oceniania. Ale także można śmiało po prostu korzystać ze strony i najpotężniejszych modeli AI za darmo. Należy jednak pamiętać że wszystkie wiadomości na stronie są publicznie dostępne, a wielkie firmy wykorzystują je do trenowania modeli. ”

“The pricing difference here is what’s absolutely mind-blowing. Kimi K3 is $3/$15 compared to Claude Fable’s eye-watering $10/$50. That’s a 3x+ cost reduction for better performance on agentic webdev. Even GLM 5.2 at $1.40/$4.40 sitting at Rank 4 is an insane value proposition. It really shows that the real existential danger the Western frontier labs were warning us about was actually just market competition. When you gatekeep your models for months under the guise of national security and safet”

“Arena.ai which everyone knew as LMArena or Chatbot Arena just got a fresh new name. What started as a small UC Berkeley PhD research project comparing language models has grown into one of the biggest community platforms where people vote on real tasks like coding reasoning writing and creative work. Main changes worth noting New simple name Arena Updated look with pillar style logo scholastic colors and clean academic fonts Website now lives at arena.ai Video evaluations already working on the ”

“The worst AI page ever. The number of bugs, errors, and annoying things that have been happening in recent months is so ridiculous that even with all the time in the universe, I wouldn't have time to count them all. The limits have been lowered, models have been removed without any notice on the page, you have to constantly do captchas over and over when you make more than 3 consecutive requests The only reason I still use this site is because of its models. ”

“how to control GLM 5.2 max medium or low in opencode ?I only have GLM 5.2 !I have an API key from Z ai also in the opencode go and both I cannot control thier effort level”

“Some benchmaxxing going on there with GLM since (as a heavy daily user of GLM 5.1 ) it's definitely a step below both Opus (as per the chart) and also GPT 5.4-High (via Codex CLI) in my experience. But, it's better than both Sonnet and GPT 5.4-Medium . That's an achievement.And its pleasant to chat with (not too concise, not too verbose) whilst doing OK with UI/front-end; something that GPT gets wrong imo. OpenWeights turns out not to be a red herring after all...”