LMArena Officially Launches, Dedicated to Providing a Neutral AI Evaluation Platform

AIbase基地

Published inAI News · 4 min read · Apr 18, 2025

The highly regarded AI evaluation platform, LMArena, recently announced the formation of a new company, Arena Intelligence Inc., to provide stronger resources for future project improvements. In a blog post, the LMArena founding team explained that this new company will allow them to enhance the capabilities of their large language model (LLM) testing platform while maintaining its neutrality, aiming to provide AI users with a fair evaluation environment free from corporate influence.

LMArena, founded in 2023 by researchers at UC Berkeley, has quickly become one of the most respected AI benchmarking platforms in the industry. It has partnered with several prominent companies, including Google, OpenAI, and Anthropic, becoming a crucial tool for the AI community in evaluating models. Initially, LMArena was primarily funded through donations and grants, including support from venture capital firm Andreessen Horowitz, Google's Kaggle data science platform, and Together Computer Inc.

In their blog post, the LMArena team noted that some members recently graduated from UC Berkeley and wished to continue their involvement in the project. They aim to better serve the AI community while upholding LMArena's original principles of neutrality and openness. They emphasized: "Our rankings will never favor any provider, but will genuinely reflect community preferences—this is the core of our design. Maintaining neutrality and earning community trust will always be key to our success."

Following the establishment of the new company, the founding team stated they have begun rebuilding core functionalities of the LMArena platform, with the new version currently undergoing beta testing and welcoming community feedback. They have addressed several bugs, improved user experience, and added new features such as login, chat history, and personalized leaderboards. Furthermore, the new company will support more open research, focusing on new evaluation projects like WebDev Arena, RepoChat Arena, and Search Arena.

While the founding team didn't disclose detailed monetization plans, they stated that their business model is not yet fully defined and they haven't secured any funding. However, they are confident in their future profitability, believing that AI companies urgently need neutral and reliable evaluation services to accelerate model development and improve real-world performance.

Chatbot Arena, AI Benchmarking Platform, Launches New Company

Amidst the rapid growth of the AI industry, Chatbot Arena, a crowdsourced AI benchmarking project, is expanding its reach by officially launching a new company, Arena Intelligence Inc. According to Bloomberg, Chatbot Arena aims to leverage this new entity to secure more resources, significantly enhancing the platform's functionality and services. Founded in 2023, Chatbot Arena is primarily spearheaded by the University of California, Berkeley...

Perplexity's Sonar Challenges Google Gemini's Search Dominance

In the latest LM Arena Search Arena evaluation, Perplexity's Sonar-Reasoning-Pro-High model performed exceptionally well, tying with Google's Gem-2.5-Pro-Grounding model for first place and achieving a 53% win rate in head-to-head comparisons. This news has sent shockwaves through the search engine industry, showcasing Perplexity's powerful AI search technology.

Meta's Llama-4-Maverick Plummets in Rankings, Raising Concerns of Benchmark Manipulation

Meta's open-source large language model, Llama-4-Maverick, has experienced a dramatic drop in LMArena rankings, plummeting from second place to 32nd. This significant shift has sparked widespread skepticism among developers, who suspect Meta may have manipulated the benchmark by submitting a specially optimized version. The issue stems from Meta's April 6th release of its latest large language model, Llama 4, encompassing three versions: Scout, Maverick, and Behemoth.

Meta Accused of AI Model Double Standard: Maverick's Performance Varies Widely Between Evaluation and Public Versions

Meta released its new flagship AI model, Maverick, on Saturday. The model ranked second in the LM Arena benchmark. LM Arena is a testing platform that relies on human raters to compare different model outputs and select their preferences. However, several AI researchers quickly discovered that the version of Maverick deployed to LM Arena appears significantly different from the version widely used by developers. Meta acknowledged in its announcement that the Maverick on LM Arena is an experimental version.

Breaking Free from Stiff AI: Midjourney and NYU Unlock New Dimensions in Creative Text Generation, Diversity Soars by 23%!

Researchers from Midjourney and New York University have collaborated on a novel approach to significantly enhance the diversity of creative text generated by language models while minimizing quality loss. Detailed in a recent research paper, this technique centers on incorporating a 'deviation metric' into the AI's training process. It works by quantifying the difference between each generated text and other texts created for the same prompt. Researchers utilize text embeddings and their pairwise cosine distances to calculate these differences, thereby providing the system with...