内容来源：https://techcrunch.com/2026/04/07/i-cant-help-rooting-for-tiny-open-source-ai-model-maker-arcee/

内容总结：

美国初创公司Arcee近日发布其新一代开源大语言模型“Trinity Large Thinking”。这家仅26人的团队以2000万美元的有限预算，成功开发出参数量达4000亿的模型。公司首席执行官马克·麦奎德向科技媒体TechCrunch表示，这是“非中国企业发布的能力最强的开源模型”。

该公司的目标是为欧美企业提供无需依赖中国模型的技术选择。尽管中国模型能力出众，但部分西方企业对其数据安全与地缘政治风险存在顾虑。Arcee的模型支持企业本地部署并自定义训练，同时提供云端API调用服务。

与Anthropic、OpenAI等巨头的闭源模型相比，Arcee虽在性能上未实现超越，但其开源特性避免了用户受制于商业公司的政策突变。例如，近期Anthropic突然调整政策，导致开源AI工具OpenClaw用户需额外支付费用，而Arcee模型已成为OpenClaw平台使用率最高的模型之一。

基准测试显示，Trinity Large Thinking的性能可与当前主流开源模型媲美。虽然尚无法撼动Meta公司Llama 4在开源领域的领先地位，但Arcee所有模型均采用业界公认最开放的开源协议Apache 2.0，避免了Meta模型存在的许可争议。

在开源模型创业浪潮中，Arcee展现了初创企业的创新活力。其发展路径不仅为中小企业提供了技术新选择，也为开源生态的多元化注入了新动力。

中文翻译：

美国初创公司Arcee规模虽小，仅26名员工，却以2000万美元的微薄预算打造出参数量高达4000亿的开源大语言模型。近日该公司发布了全新的推理模型。首席执行官马克·麦奎德向科技媒体TechCrunch宣称，这款名为"三位一体·宏思"的模型堪称"非中国企业发布过的性能最强的开放权重模型"。

这番表态透露出Arcee的目标——为欧美企业提供无需选用中国模型的替代方案，这个愿景令人不禁为之喝彩。尽管中国模型性能卓越，但其潜在风险仍令西方世界担忧：这些技术可能将权力乃至数据置于意识形态相左的政府掌控之中。

通过Arcee的方案，企业既可下载模型并根据自身需求进行训练与本地部署，也能选用其云端托管版本并通过API接口调用。虽然Arcee的模型性能尚未超越Anthropic或OpenAI等巨头的闭源模型，但用户也无需受制于这些行业巨擘的反复无常。

以编码能力出众的Claude模型为例，它本是开源AI智能体工具OpenClaw用户的热门选择。然而Anthropic上周突然宣布其订阅服务将不再支持OpenClaw调用，用户需额外付费使用。（今年2月，OpenClaw创始人彼得·斯坦伯格已宣布加入Anthropic的最大竞争对手OpenAI。）相比之下，麦奎德自豪地指出，OpenRouter平台数据显示其模型已成为OpenClaw用户最青睐的选项之一。

限时特惠：立省482美元参与2026科技风暴大会
优惠截止：太平洋时间4月10日晚11:59
寻找您的下一轮融资、下一名精英、下一个突破机遇。2026科技风暴大会汇聚上万名创业者、投资者与科技领袖，在为期三天的盛会中呈现250多场实战研讨、高效人脉对接与定义市场的创新成果。立即注册锁定优惠。

那么"三位一体·宏思"的实际表现如何？根据该公司向TechCrunch提供的基准测试结果，其性能可与其他顶尖开源模型比肩。正如我们此前报道所言，这款模型虽未对美国开源模型的领军者——Meta的Llama 4构成直接威胁，但也规避了Meta模型那种暧昧不清的"伪开源"许可协议问题。Arcee所有"三位一体"系列模型均采用开源许可的黄金标准Apache 2.0协议发布。

需要说明的是，美国还有无数其他初创公司提供开源模型。作为初创企业创新精神的拥护者，我也同样为它们加油助威。

英文来源：

Arcee, a tiny 26-person U.S. startup that built a massive, 400B-parameter open source LLM on a $20 million shoestring budget, has released its new reasoning model. Arcee calls the model Trinity Large Thinking — and it’s the most capable open-weight model “ever released by a non-Chinese company,” claims CEO Mark McQuade to TechCrunch.
As that comment implies, Arcee has a goal that I can’t help but root for: It wants to give U.S. and Western companies a model that gives them no reason to use a Chinese-based one.
While Chinese models are extremely capable, they are perceived as risky, putting power, and perhaps data, into the hands of a government that doesn’t share all of the Western world’s ideals.
With Arcee, companies can download the model, train it to their own needs, and use it on premises. Companies can also use Arcee’s cloud-hosted version, accessible via API.
While Arcee’s models are not outperforming the closed source models from the big labs like Anthropic or OpenAI, they’re not being held hostage by the whims of those giants, either.
For instance, Claude, with its exceptional abilities to code, has been a popular choice for users of open source AI agent tool OpenClaw. But Anthropic pulled the rug out from them last week when it told users that their Anthropic subscriptions will no longer cover OpenClaw usage — they will have to pay additionally for that. (In February, OpenClaw creator Peter Steinberger said he was joining Anthropic’s biggest rival, OpenAI.)
In contrast, McQuade proudly points to data from OpenRouter that says it has become one of the top models used with OpenClaw.
This Week Only: Up to $482 savings for Disrupt 2026
Offer ends April 10, 11:59 p.m. PT
Your next round. Your next hire. Your next breakout opportunity. Find it at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, powerful introductions, and market-defining innovation. Register now to secure these savings.
This Week Only: Up to $482 savings for Disrupt 2026
Offer ends April 10, 11:59 p.m. PT
Your next round. Your next hire. Your next breakout opportunity. Find it at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, powerful introductions, and market-defining innovation. Register now to secure these savings.
So, how good is Trinity Large Thinking? It is comparable to some of the other top open source models, according to the benchmark results it shared with TechCrunch.
As we previously reported, it is not a head-to-head threat to the big cheese among U.S.-built open models: Meta’s Llama 4. But it also doesn’t have the odd, not-really open source license issues of Meta’s model. All of Arcee’s Trinity models are released under the gold standard for OS licenses, Apache 2.0.
Just to be clear, there are also countless other U.S. startups offering open source models and, as a fan of the ingenuity of startups, I’m rooting for them, too.