内容来源：https://aibusiness.com/generative-ai/anthropic-tries-to-change-the-conversation-with-sonnet-4-6

内容总结：

谷歌云赞助发布：生成式AI应用选择指南与行业动态

生成式AI的初步应用应聚焦于提升人类信息交互体验。近日，Anthropic公司发布了其Claude模型系列的Sonnet 4.6版本，作为免费版和Pro版Claude.ai及Claude Cowork的默认模型。此次更新重点提升了模型的编码能力，在一致性、计算机使用技能和指令遵循方面均有增强，并能更好地理解上下文后进行代码修改。分析指出，此次升级标志着Anthropic正从基础模型提供商向能自主执行多步骤工作流的智能体解决方案供应商转型。

然而，此次技术发布被近期一场公关争议所笼罩。此前，一个最初支持Claude模型、名为OpenClaw（原名Clawdbot）的开源智能体框架，因被Anthropic要求更名而引发混乱。更名过程不仅给用户带来了安全风险，也为竞争对手OpenAI创造了机会——后者迅速招募了该框架的核心开发者。行业分析师认为，这一事件严重损害了Anthropic致力于打造的企业级友好形象，甚至冲淡了其技术更新的市场关注度。

尽管面临争议，Anthropic通过Sonnet 4.6展示了其战略转向：将上下文窗口扩展至百万令牌，支持对完整代码库或法律档案进行处理；在开发者平台上提供自适应与扩展思考能力，以平衡推理深度与成本效率；并通过MCP连接器与Excel等平台集成，确保与企业数据的安全交互。分析师总结道，Anthropic正明确朝着企业级AI应用与平台公司的方向演进，尽管其能力构建尚未完成，但战略路径已清晰显现。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI应用场景
要开始使用生成式AI，首先应关注能够改善人类信息交互体验的领域。

尽管此次更新展现了该厂商试图超越单纯模型供应商的定位，但其发布会被充满争议的智能体新闻所掩盖。Anthropic推出的Sonnet 4.6模型更新彰显了该公司对企业级AI的承诺，尽管其管理开源智能体框架的方式引发争议——该框架最初是为支持其Claude基础模型而构建的。

2月17日，这家模型制造商将Sonnet 4.6作为默认模型引入生成式AI模型的网页版Claude.ai，并向免费版和专业版用户开放Claude Cowork功能。Anthropic表示，新版本增强了编码能力，在一致性、计算机使用技能和指令遵循方面均有提升。该公司指出，虽然该模型在计算机使用方面尚不及人类，但其进步速度持续加快，且在修改代码前理解上下文的能力突出。

Sonnet 4.6的发布适逢上周末的新闻：OpenAI聘用了Peter Steinberger，他是近期流行的开源框架OpenClaw的开发者，该框架允许用户在自己的硬件上构建个人AI智能体。数月前刚推出时，OpenClaw（原名Clawdbot）曾引导用户使用Claude Opus，并建议他们通过Claude 3.5运行项目。然而，Anthropic要求Steinberger停止使用Claude相关名称，迫使其先将名称改为Moltbot，最终定为OpenClaw。这次更名需求让诈骗者有机可乘，劫持了OpenClaw的账户。此外，这也为OpenAI趁机聘用Steinberger创造了条件，不过OpenClaw将作为OpenAI基金会的一部分保持开源。

Futurum Group分析师David Nicholson表示："OpenClaw事件危及Anthropic作为企业友好型供应商的声誉，至少在本周，它让该公司在Sonnet方面的任何进展都黯然失色。Anthropic原本正在开辟作为企业AI合作伙伴的空间，以区别于OpenAI。但可以说，Anthropic浪费了所有这些优势。"他补充道，尽管该公司鼓吹智能体AI的必要性，却未能认识到像OpenClaw这样的框架可能为其服务消费带来的机遇。

Nicholson还指出，Anthropic强迫OpenClaw从Clawdbot更名的立场，使OpenClaw用户面临更大的安全风险。因此，尽管Sonnet 4.6的发布是模型改进的又一例证，但对企业的意义有限。"Anthropic本周即便发布了通用人工智能（AGI），这条新闻仍会被OpenClaw的公关灾难所掩盖。"

不过，Gartner分析师Arun Chandrasekaran认为，通过Sonnet 4.6，Anthropic正试图证明自己不再仅仅是模型制造商。"Claude 4.6（Opus和Sonnet）标志着Anthropic从模型供应商向智能体解决方案的转型，使模型能够以较高准确度自主执行多步骤业务工作流。"他补充说，通过将上下文窗口扩展到100万tokens，更新后的模型让企业能够通过单次提示处理整个代码库或法律档案。

此外，Sonnet 4.6在Claude开发者平台上支持自适应和扩展思维，使企业能够平衡"高阶推理与成本效益"。Chandrasekaran指出，与Excel等平台的MCP连接器集成也确保了Sonnet 4.6能够安全地与企业数据交互。"Anthropic正明显从模型供应商转向企业AI应用和平台公司。虽然他们尚未具备所有必要能力，但这是他们既定的战略方向。"

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
While the update shows how the vendor seeks to be more than just a model provider, the release was overshadowed by controversial agentic news.
Anthropic's launch of its Sonnet 4.6 model update hones the vendor's allegiance to enterprise AI, despite controversy over how it managed an open source agentic framework that initially supported its Claude foundation model.
On Feb. 17, the model maker introduced Sonnet 4.6 as the default model in Claude.ai, the web version of the generative AI model, and Claude Cowork for free and pro users. The updated version boasts improved coding skills, with enhancements in consistency, computer use skills, and instruction-following, according to Anthropic. The vendor said that while the model is not as good as humans in computer use, its rate of progress continues to accelerate. The model is also good at reading context before modifying code.
The release of Sonnet 4.6 comes in the wake of news last weekend that OpenAI has hired Peter Steinberger, the developer behind OpenClaw, a recently popular open source framework that enables users to build personal AI agents on their own hardware. At its launch a few months ago, OpenClaw (originally named Clawdbot) directed users to Claude Opus, urging them to run their projects with Claude 3.5. However, Anthropic asked Steinberger to stop using the Claude takeoff name, forcing him to change it to Moltbot before he settled on OpenClaw. The need to rebrand gave scammers a chance to hijack OpenClaw’s account. Moreover, it also allowed OpenAI to swoop in and hire Steinberger,, although OpenClaw will remain open source as part of a foundation at OpenAI.
The situation with OpenClaw jeopardizes Anthropic's reputation as an enterprise-friendly vendor and -- at least this week -- overshadows any development the vendor has made with Sonnet, said David Nicholson, an analyst at Futurum Group.
"Anthropic was carving out a space as the enterprise AI partner to partner with over OpenAI. I would say Anthropic squandered all of that," Nicholson said. He added that while the vendor touted agentic AI as necessary, it failed to recognize the opportunity that a framework like OpenClaw could provide to drive consumption of its services.
Moreover, the stance Anthropic took by forcing OpenClaw to change its name from Clawdbot made OpenClaw’s users more vulnerable to security risks, Nicholson said.
Therefore, while the Sonnet 4.6 release is another example of models improving, for enterprises, it does not matter, Nicholson said.
"Anthropic could have released AGI [artificial general intelligence] this week, and that news would still be overshadowed by the OpenClaw PR disaster," he said.
Nevertheless, with Sonnet 4.6, Anthropic is trying to prove that it is no longer just a model maker, said Arun Chandrasekaran, an analyst at Gartner.
"Claude 4.6 (Opus and Sonnet) marks a transition from a model provider to an agentic solution for Anthropic, enabling the model to autonomously execute multi-step business workflows with good accuracy," Chandrasekaran said. He added that by expanding its context window to one million tokens, the updated model enables enterprises to process entire codebases or legal archives with a single prompt.
Moreover, Sonnet 4.6's support for adaptive and extended thinking on the Claude Developer platform enables enterprises to balance "high-level reasoning with cost efficiency," Chandrasekaran added.
Integration with MCP connectors for platforms such as Excel also assures enterprises that Sonnet 4.6 can safely interact with their enterprise data, he said.
"Anthropic is clearly transitioning from a model provider to an enterprise AI applications and platform company," Chandrasekaran said. "While they don't have all the required capabilities yet, it is the strategic direction they are headed."