内容来源：https://aibusiness.com/agentic-ai/mistral-s-model-vibe-long-running-code-in-the-cloud

内容总结：

谷歌云专题报道：企业如何选择首个生成式AI应用场景？Mistral最新动作给出答案

在生成式AI浪潮中，企业如何迈出第一步？谷歌云赞助的分析指出，应优先聚焦于能改善人类信息交互体验的领域。法国AI初创公司Mistral的策略正是如此——通过强化自然语言交互，让编程更易上手，同时兼容现有代码库。

随着“氛围编程”（vibe coding）概念日渐流行，Mistral AI在其AI聊天机器人中推出了远程编程代理“Vibe”及“工作模式”（Work Mode），均由全新模型驱动。4月29日，Mistral宣布其云端编程代理可在无监督情况下长时间独立运行。Vibe中的远程代理基于Mistral Medium 3.5模型，这是该法国公司新推出的开源权重默认模型，拥有1280亿参数和256K上下文窗口，专注于指令遵循、推理与编程任务。

这一更新标志着“氛围编程”在过去一年的迅猛发展。自OpenAI联合创始人安德烈·卡帕斯于2025年提出该概念后，用自然语言通过AI模型编程的热度飙升。Mistral已全面押注此赛道，去年推出Mistral Vibe CLI命令行界面，允许开发者用英文描述任务；今年初更推出Vibe 2.0，使代理能在指令模糊时主动要求澄清。

Forrester Research分析师威廉·麦基翁-怀特指出，Mistral新模型旨在“让开发者用更少投入做更多事情”。与Anthropic的Claude Code拥有独立界面不同，Mistral的远程编程代理可直接连接代码仓库，无论是通过网页还是本地操作，都能保持工作上下文连贯，方便从研究到代码工作流的提示生成。

他同时表示，虽然Medium 3.5在编程或企业部署代理方面未必优于其他竞品，但其优势在于“可同时运行大量编程代理”，让企业开发者能长时间并行执行多项任务。不过他也质疑Mistral如何维持系统长期运行，特别是“多会话时的内存管理和模型上下文追踪”。

此外，Mistral还在Le Chat中推出了“工作模式”预览版，由新编排层和Medium 3.5模型驱动，支持读写及同时调用多种工具。用户可通过该模式在聊天界面处理邮件、消息和日历事项，还能跨网页搜索后生成可编辑的简报或报告。

分析人士苏认为，尽管OpenAI通过Agent SDK提供了类似体验，但Mistral“试图提供更企业级、对企业更友好的替代方案”。他补充说，Mistral将用户熟悉的聊天界面与管理代理、模型及任务路由的编排层结合，构成了“极具吸引力的价值主张”。这种“服务打包”策略同时覆盖了消费者和企业市场，是其竞争对手尚未做到的。尽管Mistral在欧美已具知名度，但在亚太市场仍相对陌生。

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，首先应聚焦于能够改善人类与信息交互体验的领域。

Mistral的方法强调自然语言交互，让编程更易上手，同时支持与现有代码库的集成。

随着“氛围编程”日益流行，Mistral AI在Vibe中引入了远程编码代理，并在其AI聊天机器人中新增了工作模式，所有这些均由新模型驱动。

Mistral于4月29日透露，其云端编码代理可长时间独立运行，无需人工监控。Vibe中的远程代理由Mistral Medium 3.5驱动，这是这家法国初创公司新推出的开源权重默认模型，同样驱动着Mistral Vibe和Le Chat这两款编码代理。Mistral表示，该模型专为长时间稳定执行编码与生产力任务而设计。它拥有128B参数和256K上下文窗口，能够处理指令遵循、推理和编码等任务。

Mistral的远程代理和模型更新，体现了过去一年间“氛围编程”的迅猛发展。自OpenAI联合创始人安德烈·卡帕西在2025年提出这一术语以来，使用自然语言通过AI模型编程的热潮便一触即发。Mistral已全面投入其中，发布了多款优先采用自然语言而非逐行编码的模型与工具。去年，这家总部位于巴黎的供应商推出了Mistral Vibe CLI（命令行界面），开发者可用英语描述任务。今年早些时候，它又发布了Vibe 2.0，当指令不清晰时，代理可主动要求澄清。

Forrester Research分析师威廉·麦基翁-怀特表示，借助新模型，Mistral旨在让开发者更轻松地“用更少资源完成更多工作”。

“这种‘轻松化’也让你能通过任意界面做更多事，”麦基翁-怀特补充道。他指出，借助该模型和远程氛围编码，开发者可以直接访问与代码库相连的编码终端或界面，无论是通过网页还是本地环境。这与Anthropic不同——Claude Code拥有独立于Claude本身的单独界面。

“Mistral有助于让工作保持上下文连贯，使得从研究到编码工作流的提示更加便捷，同时仍支持在CLI中交互，”麦基翁-怀特说。

尽管企业需要Mistral带来的连接性，但麦基翁-怀特认为，Medium 3.5在编码或帮助企业创建和部署编码代理方面，未必优于市场上其他模型。

“但你可以轻松运行比大多数竞争对手更多的编码代理，”他说。他补充道，借助远程氛围编码代理，企业开发者可让其长时间运行，从而并行执行更多任务，而非串行处理。

然而，麦基翁-怀特表示，Mistral计划如何保持系统持续运行仍不明朗。

“我最大的疑问在于长期记忆和模型上下文，以及他们计划如何帮助系统跨多个会话追踪用户行为，”他说。

Mistral还在Le Chat中推出了名为“工作模式”的新代理模型。该模式以预览形式发布，由新框架和Mistral Medium 3.5驱动。它使Le Chat能够同时读写并使用多种工具。这意味着用户可以通过工作模式在Le Chat中处理邮件、消息和日历。用户还能跨网页研究主题，生成简报或报告，并在发送前进行编辑。

虽然OpenAI通过其Agent SDK和可连接ChatGPT的连接器提供了类似体验，但苏认为：“Mistral正试图提供一种替代方案，且更偏向企业级或企业就绪。”

他补充道，AI聊天机器人一直是企业信任的痛点。然而，Mistral采用用户熟悉的聊天界面，并辅以管理代理和模型、将其导向正确任务的编排层，这一方法“提供了相当有吸引力的价值主张”。

“服务的捆绑既扩大了他们对消费者市场的吸引力，也增强了对企业市场的吸引力，这未必是其竞争对手已做到的，”苏说。他补充道，该供应商在欧洲和美国知名度较高，但在亚太市场仍相对鲜为人知。

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.
Mistral's approach prioritizes natural language interactions, making coding more accessible while allowing for integration with existing code repositories.
With vibe coding growing in popularity, Mistral AI introduced remote coding agents in Vibe and a Work mode in its AI chatbot, all powered by a new model.
Mistral revealed on April 29 that its cloud coding agents can run for extended periods independently without supervision. The remote agents in Vibe are powered by Mistral Medium 3.5, the French startup’s new open-weight default model, as are the coding agents Mistral Vibe and Le Chat. The model is built to run for long stretches on coding and productivity, Mistral said. It has 128B parameters with a 256K context window, which allows it to manage instruction-following, reasoning and coding.
Mistral’s remote agents and model update exemplify the rapid growth of vibe coding in just the past year. The popularity of coding with AI models, using natural language, has exploded since OpenAI co-founder Andrej Karpathy coined the term in 2025. Mistral has moved heavily into it with the release of various models and tools that prioritize natural language over line-by-line coding. Last year, the Paris-based vendor introduced Mistral Vibe CLI, its command-line interface that developers can use to describe tasks in English. It also released Vibe 2.0 earlier this year, which enables agents to ask for clarification when a prompt is unclear.
With the new model, Mistral is seeking to make it easier for developers to do more with less, said William McKeon-White, an analyst at Forrester Research.
“This 'making it easier' also allows you to do more from whatever interface you want,” McKeon-White said. He added that with the model and remote-vibe coding, developers can access a coding terminal or interface directly connected to a code repository, either using the web or locally. This differs from Anthropic in that Claude Code has a separate interface, distinct from Claude itself.
“Mistral helps keep work more in context and allows for easier prompts to research to code workflows, while still allowing you to interact in a CLI,” McKeon-White said.
While enterprises need the connectivity that Mistral is bringing, Medium 3.5 is not necessarily better than other models on the market when it comes to coding or helping enterprises to create and deploy coding agents, McKeon-White continued.
"But you can easily run many more [coding agents] than most other competitors,” he said. He added that with the remote vibe coding agents, enterprise developers can leave them running for an extended period , enabling them to perform more tasks in parallel rather than in sequence.
However, how Mistral plans to keep its system running is still unclear, McKeon-White said.
“My big questions are around long-running memory and model context and how they plan to help their system keep track of what people are doing over multiple sessions,” he said.
Mistral also rolled out a new agentic model, called Work Mode, in Le Chat. Introduced in preview, Work Mode is powered by a new harness and Mistral Medium 3.5. It enables Le Chat to read/write and use multiple tools at once. This means that users can catch up on emails, messages and calendars all in Le Chat through Work Mode. Users can also research topics across the web and then produce a brief or report that can be edited before sending, Mistral said.
While OpenAI provides a similar experience with its Agents SDK and connectors that can be connected to ChatGPT, “Mistral is trying to provide an alternative but something that is more enterprise-grade or enterprise-ready,” Su said.
He added that the AI chatbot has always been a source of mistrust for enterprises. Still, the Mistral approach of using a chat interface people are used to, with an orchestration layer that manages agents, models and routes them to the right task, is a "pretty compelling value proposition," he continued.
“The bundling of the services does broaden their appeal to both the consumer and enterprise markets, which is not necessarily something that their competitors have done,” Su said. He added that the vendor is well known in Europe and U.S., but still relatively unknown in the APAC market.