广告
加载中

美团LongCat-2.5发布:AI向主业纵深 强化商业分析等能力

姜琪 2026-11-02 11:34
姜琪 2026/11/02 11:34

邦小白快读

EN
全文速览

总:美团11月2日发布新一代大模型LongCat-2.5,这是一款面向长程智能体任务的原生多模态模型。

1.关键参数:总参数1.6万亿,训练全程由国产算力完成,整体能力比肩当前国际成熟闭源模型和头部开源模型。

2.能力升级:在6月发布的LongCat-2.0基础上,进一步强化原生视觉理解与智能体编程能力,并优化国产算力上的长上下文训练与推理效率。

3.实操案例:在品牌多渠道经营分析中,模型能自主构建渠道利润表,识别出支付GMV增长23.2%而渠道利润下降1.7%的增长质量问题,并结合清货经济价值、用户复购与库存风险提出差异化资源配置建议。

4.可用工具:美团推出AI经营工作台“妙手”,已服务古茗、徐升记、益禾堂等连锁品牌,覆盖品牌经营分析、门店管理和增长获客场景。

总:文章显示AI正直接嵌入品牌的多渠道经营、门店管理与获客场景。

1.渠道与价格:案例中模型自主构建渠道利润表,识别出支付GMV增长23.2%而渠道利润下降1.7%的增长质量问题,提示品牌警惕只增GMV不增利润的渠道扩张。

2.资源配置:模型结合清货经济价值、用户复购与库存风险,提出差异化资源配置建议,完成从经营核算、问题诊断到决策建议的完整分析。

3.品牌规则与指标:AI经营工作台“妙手”帮助商家理解行业指标和品牌规则,并在商家授权范围内调用工具与系统、完成经营任务执行。

4.落地样本:妙手已服务古茗、徐升记、益禾堂等连锁品牌,融入品牌经营分析、门店管理和增长获客等高价值经营场景。

总:文章释放的核心信号是AI工具化落地,商家有可直接借用的经营帮手,同时也有增长质量的提醒。

1.机会提示:美团基于LongCat系列模型推出AI经营工作台“妙手”,帮助商家在高价值经营场景中理解行业指标和品牌规则,并在授权范围内调用工具与系统完成经营任务。

2.可学习点:模型能自主核算渠道利润表,再诊断问题、给出差异化资源配置建议,商家可参照这条从发现问题到决策执行的路径。

3.风险提示:案例中支付GMV增长23.2%而渠道利润下降1.7%,说明增长质量存在问题,需同时关注清货经济价值、用户复购与库存风险。

4.合作方式与前提:工具在商家授权范围内调用工具与系统,数据授权是使用前提;古茗、徐升记、益禾堂等服务案例可作为参考。

总:文章的国产算力进展与智能体编程能力,对制造企业的数字化升级有参考价值。

1.国产算力可行性:LongCat-2.0已于6月在国产算力上完成万亿参数模型的全流程训练与推理,LongCat-2.5训练全程由国产算力完成,说明国产算力能够支撑大模型规模化训练。

2.技术能力:新模型是面向长程智能体任务的原生多模态模型,总参数1.6万亿,强化原生视觉理解与智能体编程能力,并优化国产算力上的长上下文训练与推理效率。

3.数字化启示:美团强调AI要真正进入真实工作流、解决真实问题,并形成长期稳定的基础设施能力,不做Token工厂,模型用于支持主业。

4.场景参照:模型重点强化商业分析、软件开发等能力,为本地生活商家配备AI帮手,并已在连锁品牌的经营分析、门店管理与增长获客中落地。

总:文章给出行业趋势判断与可复用的解决方案范式。

1.行业趋势:国产算力完成万亿参数模型全流程训练与推理,国芯国模更进一步,模型原生多模态、面向长程智能体任务,整体能力比肩国际成熟闭源模型和头部开源模型。

2.新技术:1.6万亿参数、原生视觉理解、智能体编程,以及国产算力上的长上下文训练与推理效率优化。

3.客户痛点:品牌多渠道经营中支付GMV增长23.2%而渠道利润下降1.7%的增长质量问题,以及清货经济价值、用户复购与库存风险难以兼顾。

4.解决方案:模型自主构建渠道利润表,从经营核算、问题诊断走到决策建议;美团AI经营工作台“妙手”帮助商家理解行业指标和品牌规则,在授权范围内调用工具与系统执行任务,已服务古茗、徐升记、益禾堂。

总:美团对AI的定位与平台化做法,为平台运营与生态管理提供参照。

1.平台定位:CEO王兴明确不会成为Token工厂,模型和AI产品用于支持主业,提升用户、商户体验和公司经营效率;AI的长期价值在于真正进入真实工作流、解决真实问题,形成长期稳定的基础设施能力。

2.平台做法:基于LongCat系列模型推出AI经营工作台“妙手”,帮助商家理解行业指标和品牌规则,并在商家授权范围内调用工具与系统、完成经营任务执行。

3.生态与边界:文章明确商家授权范围是调用工具与系统的前提,涉及商家数据与系统调用的权限管理。

4.运营管理:妙手已服务古茗、徐升记、益禾堂等连锁品牌,覆盖品牌经营分析、门店管理和增长获客,平台AI正从发现问题走向决策和执行。

总:文章呈现出国产大模型与本地生活主业深度结合的产业新动向。

1.技术动向:LongCat-2.0于6月在国产算力上完成万亿参数模型全流程训练与推理;LongCat-2.5总参数1.6万亿,面向长程智能体任务、原生多模态,训练全程由国产算力完成,能力比肩国际成熟闭源模型和头部开源模型。

2.商业模式:不以Token售卖为目标,模型与AI产品用于支持主业;推出AI经营工作台“妙手”,在品牌经营分析、门店管理、增长获客等场景服务连锁品牌。

3.新问题:案例中支付GMV增长23.2%与渠道利润下降1.7%形成背离,构成增长质量议题,经营建议涉及清货经济价值、用户复购与库存风险。

4.治理启示:明确在商家授权范围内调用工具与系统,数据与系统调用边界值得进一步研究;模型价值需以能否解决真实问题来检验。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

Summary: Meituan released LongCat-2.5 on November 2, a native multimodal model designed for long-horizon agentic tasks.

1. Key parameters: 1.6 trillion total parameters; the entire training process was completed on domestic compute; overall capability is on par with mature international closed-source models and leading open-source models.

2. Capability upgrades: Building on LongCat-2.0, released in June, it further strengthens native visual understanding and agentic coding, while optimizing long-context training and inference efficiency on domestic compute.

3. Practical case: In a multi-channel brand operations analysis, the model independently built a channel profit-and-loss statement and identified a growth-quality problem: payment GMV grew 23.2% while channel profit fell 1.7%. Combining clearance economic value, user repurchase, and inventory risk, it proposed differentiated resource-allocation recommendations.

4. Available tool: Meituan launched an AI operations workbench called Miaoshou, which has served chain brands such as Guming, Xushengji, and Yihetang, covering brand operations analysis, store management, and growth and customer acquisition scenarios.

Summary: The article shows that AI is being embedded directly into brands’ multi-channel operations, store management, and customer acquisition.

1. Channels and pricing: In the case, the model independently built a channel profit-and-loss statement and identified a growth-quality issue: payment GMV rose 23.2% while channel profit declined 1.7%, warning brands against channel expansion that increases GMV without increasing profit.

2. Resource allocation: Combining clearance economic value, user repurchase, and inventory risk, the model proposed differentiated resource-allocation recommendations, completing a full analysis from operations accounting and problem diagnosis to decision recommendations.

3. Brand rules and metrics: The AI operations workbench Miaoshou helps merchants understand industry metrics and brand rules, and—within the scope authorized by merchants—calls tools and systems to execute operational tasks.

4. Implementation examples: Miaoshou has served chain brands including Guming, Xushengji, and Yihetang, integrating into high-value operational scenarios such as brand operations analysis, store management, and growth and customer acquisition.

Summary: The core signal is the practical deployment of AI tools: merchants now have an operations assistant they can directly use, along with a reminder about growth quality.

1. Opportunity: Based on the LongCat series, Meituan launched the AI operations workbench Miaoshou to help merchants understand industry metrics and brand rules in high-value operational scenarios and, within authorized scope, call tools and systems to complete operational tasks.

2. Takeaway: The model can independently calculate channel profit-and-loss statements, then diagnose problems and offer differentiated resource-allocation recommendations. Merchants can follow this path from problem discovery to decision execution.

3. Risk warning: In the case, payment GMV grew 23.2% while channel profit fell 1.7%, indicating a growth-quality problem. Merchants need to monitor clearance economic value, user repurchase, and inventory risk at the same time.

4. Cooperation model and prerequisites: The tool calls tools and systems within the merchant’s authorized scope; data authorization is a prerequisite for use. Service cases such as Guming, Xushengji, and Yihetang can serve as references.

Summary: The article’s progress in domestic compute and agentic coding capabilities offers a reference for manufacturers’ digital upgrades.

1. Feasibility of domestic compute: LongCat-2.0 completed full-process training and inference of a trillion-parameter model on domestic compute in June, and LongCat-2.5’s entire training process was completed on domestic compute, showing that domestic compute can support large-scale model training.

2. Technical capabilities: The new model is a native multimodal model for long-horizon agentic tasks, with 1.6 trillion total parameters. It strengthens native visual understanding and agentic coding and optimizes long-context training and inference efficiency on domestic compute.

3. Digitalization implications: Meituan emphasizes that AI must truly enter real workflows, solve real problems, and become a long-term, stable infrastructure capability. It will not be a “Token factory”; models are used to support the core business.

4. Scenario reference: The model particularly strengthens business analysis and software development capabilities, equips local-services merchants with an AI assistant, and has already been deployed in chain brands’ operations analysis, store management, and growth and customer acquisition.

Summary: The article offers an industry trend assessment and a reusable solution framework.

1. Industry trend: Domestic compute has completed full-process training and inference for a trillion-parameter model, marking further progress in domestic chips and models. The model is natively multimodal and designed for long-horizon agentic tasks, with overall capability on par with mature international closed-source models and leading open-source models.

2. New technologies: 1.6 trillion parameters, native visual understanding, agentic coding, and optimized long-context training and inference efficiency on domestic compute.

3. Customer pain points: In multi-channel brand operations, payment GMV grew 23.2% while channel profit fell 1.7%, a growth-quality problem; clearance economic value, user repurchase, and inventory risk are difficult to balance.

4. Solution: The model independently builds channel profit-and-loss statements, moving from operations accounting and problem diagnosis to decision recommendations. Meituan’s AI operations workbench Miaoshou helps merchants understand industry metrics and brand rules and, within authorized scope, call tools and systems to execute tasks. It has served Guming, Xushengji, and Yihetang.

Summary: Meituan’s positioning and platform approach to AI provide a reference for platform operations and ecosystem management.

1. Platform positioning: CEO Wang Xing made clear that Meituan will not become a “Token factory”; its models and AI products are used to support the core business and improve user and merchant experiences as well as company operating efficiency. The long-term value of AI lies in truly entering real workflows, solving real problems, and forming long-term, stable infrastructure capabilities.

2. Platform approach: Based on the LongCat series, Meituan launched the AI operations workbench Miaoshou to help merchants understand industry metrics and brand rules and, within the scope authorized by merchants, call tools and systems to execute operational tasks.

3. Ecosystem and boundaries: The article makes clear that merchant authorization scope is the prerequisite for calling tools and systems, involving permission management for merchant data and system calls.

4. Operations management: Miaoshou has served chain brands including Guming, Xushengji, and Yihetang, covering brand operations analysis, store management, and growth and customer acquisition. Platform AI is moving from problem discovery to decision-making and execution.

Summary: The article reveals a new industry development trend: domestic large models are deeply integrating with the core local-services business.

1. Technology trend: LongCat-2.0 completed full-process training and inference of a trillion-parameter model on domestic compute in June. LongCat-2.5 has 1.6 trillion total parameters, is designed for long-horizon agentic tasks, is natively multimodal, and was trained entirely on domestic compute; its capability is on par with mature international closed-source models and leading open-source models.

2. Business model: Token sales are not the goal; models and AI products are used to support the core business. Meituan launched the AI operations workbench Miaoshou to serve chain brands in scenarios such as brand operations analysis, store management, and growth and customer acquisition.

3. New questions: In the case, payment GMV growth of 23.2% diverged from a 1.7% decline in channel profit, creating a growth-quality issue. Operational recommendations involve clearance economic value, user repurchase, and inventory risk.

4. Governance implications: The article specifies that tools and systems are called within the merchant’s authorized scope, and the boundaries of data and system calls merit further research. The value of models should be tested by whether they can solve real problems.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

【亿邦原创】11月2日,美团发布新一代大模型LongCat-2.5,这是一款面向长程智能体任务的原生多模态模型,总参数1.6万亿,训练全程由国产算力完成。

今年6月,美团发布LongCat-2.0,首次在国产算力上完成万亿参数模型的全流程训练与推理。在LongCat-2.0基础上,新模型进一步强化原生视觉理解与智能体编程能力,并优化国产算力上的长上下文训练与推理效率。

LongCat-2.5的发布,标志着“国芯国模”更进一步:不仅原生支持多模态,模型整体能力也比肩当前国际成熟闭源模型和头部开源模型。

今年8月28日的美团二季度财报电话会上,CEO王兴表示:AI能力的长期价值,是能否真正进入真实工作流、解决真实问题,并形成长期稳定的基础设施能力。“我们不会成为Token工厂,我们的模型和AI产品将用于支持我们的主业,提升用户、商户体验和公司经营效率。”

围绕“支持主业”这一目标,LongCat-2.5重点强化了商业分析、软件开发等能力,为本地生活商家配备AI帮手。

LongCat-2.5技术博客显示,在品牌多渠道经营分析案例中,模型能自主构建渠道利润表,识别出支付GMV(商品交易总额)增长23.2%,而渠道利润下降1.7%的增长质量问题,并结合清货经济价值、用户复购与库存风险提出差异化资源配置建议,完成从经营核算、问题诊断到决策建议的完整分析。

基于LongCat系列模型,美团推出了AI经营工作台“妙手”,帮助商家在高价值经营场景中,理解行业指标和品牌规则,并在商家授权范围内调用工具与系统、完成经营任务执行。

目前,“妙手”已经服务了古茗、徐升记、益禾堂等连锁品牌,融入了品牌经营分析、门店管理和增长获客等高价值经营场景,帮助商家从发现问题走到决策和执行。

亿邦持续追踪报道该情报,如想了解更多与本文相关信息,请扫码关注作者微信。

本文首发于 亿邦动力 官方网站

文章来源:亿邦动力

广告
微信
朋友圈

FAQ回顾

美团LongCat-2.5是什么模型?

美团LongCat-2.5是美团11月2日发布的新一代大模型,面向长程智能体任务,属于原生多模态模型,总参数1.6万亿,训练全程由国产算力完成。它在LongCat-2.0基础上强化原生视觉理解、智能体编程,并优化国产算力上的长上下文训练与推理效率。

LongCat-2.5重点强化了哪些业务能力?

LongCat-2.5重点强化商业分析、软件开发等能力,服务美团本地生活主业。技术博客案例显示,它能自主构建渠道利润表,识别支付GMV增长23.2%、渠道利润下降1.7%的增长质量问题,并结合清货经济价值、用户复购与库存风险提出差异化资源配置建议。

美团为什么强调大模型要支持主业?

美团CEO王兴在二季度财报电话会上表示,AI的长期价值在于真正进入真实工作流、解决真实问题,并形成长期稳定的基础设施能力;美团不会成为Token工厂,模型和AI产品将用于支持主业,提升用户、商户体验和公司经营效率。

美团妙手AI经营工作台是什么?

妙手是美团基于LongCat系列模型推出的AI经营工作台,帮助商家在高价值经营场景中理解行业指标和品牌规则,并在授权范围内调用工具与系统、完成经营任务执行。它已服务古茗、徐升记、益禾堂等连锁品牌,覆盖品牌经营分析、门店管理和增长获客等场景。

LongCat-2.5对本地生活商家有什么作用?

LongCat-2.5强化商业分析和软件开发能力,为本地生活商家配备AI帮手。商家可通过AI经营工作台“妙手”进行品牌经营分析、门店管理和增长获客,在授权范围内调用工具执行任务,从发现问题走到决策和执行;模型还能诊断渠道利润等经营质量问题并给出资源配置建议。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0