Here's how open-source models such as DeepSeek's are shaking up AI https://t.co/HKVZolQjTu ...

Here's how open-source models such as DeepSeek's are shaking up AI https://t.co/e9wnezT7kx ...

在性能至上的铁律下，多数人眼中"开源"的未来正面临前所未有的挑战。 Me n l o Ve n t u r e s最新发布的2 0 2 5年中期报告基于对1 5 0家以上企业及初创公司的技术决策 者调查揭示了一个令人震惊的行业转向： 当市场沉迷于开源模型的成本优势时，企业级LLM API支出却在过去六个月内从3 5亿美元 增长至8 4亿美元，实现翻倍增长，而闭源模型正在这个价值千亿的市场中建立难以逾越的 性能护城河。 这份报告的核心观点如下： 锦 秋 基 金 （ 公 众 号 ： 锦 秋 集 ；ID ： j q c a p it a l ） 认 为 ， 虽 然 Me n l o Ve n t u r e s 是 An t h r o p i c 的 投资人，可能立场上有倾向，但这篇报告在一定程度上反映了企业用户对AI产品采购的核 心考量与实际偏好，因此也做了编译。 " 企 业 级 市 场 中 ， 开 源 模 型 的 采 用 率 趋 于 平 稳 ， 因 其 性 能 仍 落 后 闭 源 模 型 9 到 1 2 个 月。" 尽 管 开 源 模 型 具 备 成 本 和 定 制 优 势 ， 但 性 能 差 距 、 部 署 ...

China is single-handedly bringing open-source models to the frontier. We now have a brand new model from ZAI called GLM4.5% that is on par with the best closed source models at reasoning, coding, and agentic capabilities. Before I tell you more about the model, let me show you some demos.This is the second model ever to successfully simulate a Rubik's cube. To be fair, I had to go back and forth a couple times with it. There was just a minor problem.And now check this out. So here's the Rubik's cube. All th ...

Model Performance & Technology - The company's models have consistently ranked among the top tier in the last two years [1] - The company possesses state-of-the-art capabilities in specific domains such as transcription and optical character recognition [1] - The company's models are optimized for local deployment, including on laptops [1] - The company is a leading actor in open-source models within the US and Europe [1] - The company adopts a full-stack approach to AI deployment, differentiating it from typical model providers [2] Enterprise Solutions & Partnerships - Enterprises often struggle to realize the P&L impact of general AI [2] - Delivering value to enterprises requires more than just models, including tooling for agent deployment, workflow orchestration, and observability [2][3] - The partnership with ENT data aims to combine technology, expertise, and talent to meet P&L expectations with AI [3]