Rachel Metz报道：Arena Intelligence Inc.——这家提供热门人工智能模型排行榜的初创公司，已筹集2亿美元融资，估值31亿美元，并计划扩展到AI安全测量领域，以满足对评估技术性能和风险日益增长的需求。此次融资将在周四公布，由Lightspeed Venture Partners和Khosla Ventures领投，Salesforce Ventures和Dell Technologies Capital参与。该初创公司在今年一月的融资轮中估值为17亿美元。
Intelligence Inc., the startup behind a popular artificial intelligence model leaderboard, has raised $200 million in funding at a $3.1 billion valuation and plans to expand into measuring AI safety, tapping into growing demand to evaluate the technology’s performance and risks. The financing, set to be announced Thursday, was led by Lightspeed Venture Partners and Khosla Ventures, with participation from Salesforce Ventures and Dell Technologies Capital. The startup was previously valued at $1.7 billion in a funding round in January.
Arena始于2023年，最初名为Chatbot Arena，是加州大学伯克利分校Sky Computing Lab的一个研究项目，允许任何人试用最先进的AI模型并对回复进行排名。在OpenAI发布ChatGPT引发AI热潮几个月后推出，Chatbot Arena很快成为行业早期采用者的热门去处，也是快速发展的AI基准测试领域的领先风向标。
Arena began in 2023 as Chatbot Arena, a research project from the University of California at Berkeley’s Sky Computing Lab that allowed anyone to try state-of-the-art AI models and rank the responses. Introduced a few months after OpenAI’s release of ChatGPT kicked off an AI frenzy, Chatbot Arena soon became a popular spot for the industry’s early adopters, and a leading indicator in the rapidly evolving field of AI benchmarking.
该学术项目于2025年转型为一家公司，此后扩展出多个不同的排行榜，让人们能在大量不同任务上衡量AI模型。据公司联合创始人兼首席执行官Anastasios Angelopoulos称，Arena现在每月吸引数千万访问者。
The academic project transformed into a company in 2025 and has since branched out to include a number of different leaderboards that let people measure AI models on aslew of different tasks. Arena now gets tens of millions of visitors per month, according to Anastasios Angelopoulos, the company’s co-founder and chief executive officer.
Arena计划于周四推出一个新排行榜，展示AI模型在多项对齐指标上的表现，对齐是一个常用术语，指软件遵循人类设定目标的程度。Angelopoulos表示，公司的“对齐指数”将追踪AI模型在编程等活动中与人类协作的有效性和安全性，通过考量模型采取未授权行动的频率、或谎称完成了实际未完成任务等信号来评估。
ArrivalArena plans to roll out a new leaderboard on Thursday showing how AI models stack up on several measures of alignment, a term often used to refer to how well the software follows the objectives people set for it. The company’s Alignment Index will track how effectively and safely AI models can work with people on activities like coding, Angelopoulos said, by considering signals such as how often a model takes an unauthorized action or tells a user it finished a task that it didn’t actually complete.
在近期发生一系列涉及失控AI系统的黑客事件后，OpenAI和Anthropic PBC等领先AI模型制造商目前正努力应对人们对其最先进AI产品日益增长的担忧。
Leading AI model makers such as OpenAI and Anthropic PBC are currently grappling with rising concerns about their most advanced artificial intelligence products after a spate of recent hacking incidents involving rogue AI systems.
安全漏洞以及行业内关于生存风险的警告不断升级，引发了对加强技术防护和监管的呼声。
The security breaches, and escalating warnings of existential risks within the industry, have sparked calls for greater safeguards and oversight of the technology.
最初，Arena的对齐指数包含二十多种不同的AI模型，其中OpenAI的GPT-6.1 Sol和Anthropic的Claude Opus 5.5分别位居第一和第二。
Initially, Arena’s alignment index includes more than two dozen different AI models, with OpenAI’s GPT-6.1 Sol and Anthropic’s Claude Opus 5.5 in first and second place, respectively.
Arena通过其Agent Arena平台从用户与AI模型的对话中提取对齐数据，该平台允许人们使用模型处理更复杂的多步骤任务。Angelopoulos表示，Arena计划随着时间的推移扩大排行榜，纳入更多与安全相关的信号。
Arena is pulling the alignment data from conversations users have with AI models through its Agent Arena, a platform that lets people use models for more complicated multistep tasks. Angelopoulos said Arena plans to expand the leaderboard to include more safety-related signals over time.
随着最新一轮融资的完成，Angelopoulos表示，Arena还计划将其评估工作扩展到更广泛的企业，帮助它们利用自身数据确定哪些AI模型适合其运行。
With the latest round of funding, Angelopoulos said Arena also plans to its expand its evaluation work to a broader range of businesses, helping them use their own data to determine which AI models make sense for them to run.
“我们目前还处于早期阶段，”Angelopoulos说。“我们已经取得了良好进展。”
“We’re in the early days of it now,” Angelopoulos said. “We’ve made good progress.”