算力交易中心 模型供应商 接入 100+ 大模型成标配,"模型超市"时代真的来了

The two sides have launched in-depth collaboration across four core dimensions: joint optimization of model inference efficiency, co-construction of large-model Token services and AI infrastructure, collaborative innovation of application paradigms, and large-scale expansion of industrial intelligent scenarios.


Unexhub × MiniMax: The “Model Supermarket” Era Has Officially Arrived

On August 5, 2026, Unexhub officially signed a strategic cooperation agreement with MiniMax. As two Shanghai-based tech leaders, Unexhub is a top-tier AI infrastructure enterprise with a 37,000P computing power scheduling capacity, while MiniMax stands as a leading large-model vendor with self-developed models and massive real-world application traffic.

The two sides have launched in-depth collaboration across four core dimensions: joint optimization of model inference efficiency, co-construction of large-model Token services and AI infrastructure, collaborative innovation of application paradigms, and large-scale expansion of industrial intelligent scenarios.

Tokens are rapidly becoming the most standardized value carrier in the AI era. As of March 2026, China’s average daily Token call volume has exceeded 140 trillion, a 1,000-fold increase compared with 2024, marking the explosive growth of the Token economy.

Nevertheless, the Token market has long suffered structural flaws in circulation and pricing. Pricing power is highly monopolized by model vendors, leaving developers and enterprises with no choice but to accept fixed rates. The same Token unit varies by up to thousands of times in price across different vendors, models and usage scenarios — a clear sign of an immature market lacking unified trading rules.

Xia Lixue, Co-founder and CEO of Unexhub, proposed transforming fragmented and heterogeneous computing power into accessible, affordable and high-performance standardized Tokens, enabling users to pay only for valid Token consumption without concerning underlying hardware and chip differences.

Essentially, Unexhub is building a professional trading exchange for the “new oil” of AI — Tokens. It upgrades traditional Token commodity transactions to standardized financial-level settlement, with four core mechanisms to reshape industrial rules:

  • Real-Time Settlement: Suppliers receive synchronous fund returns based on actual Token supply volume, eliminating cash flow pressure and long settlement cycles entirely.

  • Zero Commission for Early Settlers: The platform trades short-term profits for industrial liquidity. More settled suppliers and richer Token categories continuously strengthen the platform’s network effects and ecological barriers.

  • Market-Driven API Bidding: Breaking one-sided vendor pricing, API prices float dynamically with real supply and demand. Lower off-peak prices release suppressed long-tail demand, while higher peak prices incentivize suppliers to expand capacity, forming a healthy market cycle.

  • Full-Link Infrastructure: The platform provides complete supporting systems for merchant settlement, Token trading and fund clearing, allowing suppliers to focus solely on core Token production and supply.

Token Factory: Industrial Upgrade from “Hardware Rental” to “Token Delivery”

At WAIC 2026, Unexhub launched its groundbreaking “Front Store, Back Factory & Central Hub” strategy, completely overturning the traditional hardware rental model and realizing standardized Token mass production:

  • Central Hub — Heterogeneous Computing Dispatching Center: Centrally manages and intelligently schedules fragmented heterogeneous computing power nationwide. It currently integrates 37,000P of high-quality computing power, covering 16 mainstream chip types to maximize resource utilization.

  • Back Factory — Standardized Token Production Base: Taking power consumption and chip operating time as basic production materials, it adopts full-stack software optimization as the core production formula for high-efficiency, low-cost Token manufacturing. Over the past year, Unexhub has reduced overall inference costs to one-tenth of the original level. Leveraging the cross-cluster heterogeneous PD separation (PDD) architecture, it cuts first-Token latency by 51.5% and single-Token costs by 37.5%. According to Xia Lixue, the platform’s daily Token call volume has surged 40 times from last December to July this year, with an additional 90% cost reduction potential for future inference tasks.

  • Front Store — AI Productivity Service Portal: Packages standardized Token outputs into universal industrial services, converting technical resources into accessible, usable and scalable AI productivity for all industries.

The three-layer architecture forms a self-driven industrial flywheel. Market demand from the front store continuously optimizes Token production efficiency in the back factory, which further activates richer resource scheduling from the central hub, achieving iterative industrial upgrading.

Popularization of Model Supermarkets: Hundreds of Models Become Industry Standard Configurations

While computing power trading centers and Token factories solve industrial production and circulation problems, Model Supermarkets eliminate the pain points of complicated model selection and repeated adaptation for enterprises. Users no longer need separate docking, recharging and reconciliation for different models.

Since 2026, standardized Token operation centers and Model Supermarkets have been launched intensively across China, becoming mainstream industrial infrastructure:

Yangtze River Delta (Jiaxing) Token Operation Center: Officially launched on July 30 with an online portal. One single access enables on-demand calls to over 100 mainstream models including DeepSeek and Qwen. Adopting the “enterprise defines tasks, platform selects optimal models” logic, it automatically matches the most cost-effective model based on business scenarios. It supports five core unifications: unified API entry, unified Token metering, unified settlement, unified policy deduction, and unified security audit.

Hangzhou Model Supermarket 2.0: Built on Alibaba Cloud Bailian, one single key supports access to hundreds of mainstream models including Qwen, DeepSeek, Zhipu, Kimi, MiniMax and Doubao. One-time full reuse reduces enterprise R&D docking workload by approximately 70%.

China Mobile MoMA Platform: Gathers more than 300 open-source, closed-source and self-developed models. The Wuxi municipal “Token Supermarket” launched with over 20 commonly used enterprise-level models available for one-click calling.

One-click access to hundreds of models is no longer an exclusive feature of individual platforms, but a universal industry standard.

Core Changes Brought by the Model Supermarket Era

The widespread adoption of Model Supermarkets has completely rewritten the rules for enterprise AI application development, bringing three core changes for developers and enterprises:

1. Near-zero Model Access Cost

Traditionally, accessing a single model required reading documents, writing exclusive adaptation code and repeated debugging, with the whole process restarted for every model replacement. Now, one single access, one unified API and one universal key support free switching among over 100 models. Enterprise docking workload is cut by 70%, and business launch cycles are shortened from weeks to days.

2. Users Regain Independent Model Selection Rights

Enterprises can flexibly switch models according to actual business effects — replace DeepSeek with Kimi for better semantic understanding, or adopt MiniMax for scenario adaptation. Model switching is simplified from days of development work to a single parameter modification. The platform’s intelligent matching capability acts as a built-in “professional buyer” to select the optimal model for cost and quality balance.

3. Transparent and Controllable Full-Link Costs

Model Supermarket 2.0 equips full-dimensional Token cost comparison and optimization solutions, visualizing call volume, cost details and usage trends. Enterprises no longer need separate recharging and reconciliation, realizing unified asset management via one single account.

Engineer’s Perspective: The Underlying Logic Upgrade of the AI Industry

The strategic cooperation between Unexhub and MiniMax is far more than a simple enterprise partnership. Combined with the continuous improvement of Token factories, computing power trading centers and national Model Supermarkets, it marks a fundamental overhaul of the industry’s underlying logic.

The computing power trading center standardizes Token pricing and circulation, the Token factory optimizes Token production and cost control, and the Model Supermarket simplifies Token access and selection. The three systems jointly form a complete industrial closed loop covering production, circulation and consumption.

Unexhub’s API bidding mechanism realizes market-oriented Token pricing: individual developers enjoy ultra-low prices in off-peak periods, while suppliers gain stable revenue growth in peak demand periods. This is the fair and transparent pricing mechanism that Token, as a new-era digital currency, should possess — determined entirely by market supply and demand rather than monopoly rules.

Since the signing of the Unexhub & MiniMax cooperation on August 5, the “Model Supermarket” has bid farewell to conceptual hype. It is embodied in the 100+ model resources in Jiaxing, the one-click calling capability in Hangzhou, and the 300+ customizable models on MoMA.

The Model Supermarket era has truly arrived, with one API, one key and one access as the universal admission ticket of the industry.

As Zeng Shulin, General Manager of Unexhub, recalled, the company has grown from a 7-person startup with a 50-square-meter office to a 300+ professional team. It has joined hands with industrial partners including Mosu Space, Shanghai Yidian and Huxin Humo to build China’s first computing power ecological supermarket, committed to “maximizing Token output per unit of power and chip resources”.

When AI capabilities become as accessible, transparent and convenient as water and electricity, the Model Supermarket is no longer a marketing slogan, but a new starting point for industrial intelligent transformation. The organic integration of Unexhub’s Agentic Infra strategy, MiniMax’s massive application traffic, and the unified service capabilities of national Token operation centers has completed the leap of the Model Supermarket from concept to standard, redefining the underlying infrastructure of the global Token economy.

Unexhub Platform Core Advantages

Unexhub reshapes global Token circulation and trading rules, building a leading Token aggregation and computing power trading center. The platform gathers full-spec AI suppliers worldwide, covering mainstream large models such as Gemini, ChatGPT, Claude, DeepSeek and GLM, as well as high-quality customized models from individual developers and SMEs. Relying on multi-supplier bidding competition and millisecond-level intelligent routing, it delivers extreme cost performance.

Developers and enterprises pay strictly based on actual Token consumption, with the lowest cost as low as 30% of the official market price. AI suppliers can access the platform with one click and realize real-time settlement of call sharing revenue, achieving efficient and stable profit realization.


2026年8月5日,无问芯穹(Unexhub)与MiniMax正式签署战略合作协议。一方是拥有37000P超强算力调度能力的头部AI基础设施企业,深耕算力聚合、调度与优化;一方是自研大模型、坐拥海量真实落地场景与用户流量的顶尖模型厂商。两家扎根上海的科技企业,达成四大维度深度合作:模型推理效能体系共建优化、大模型Token服务与AI基础设施协同搭建、应用范式联合创新、行业智能化场景规模化拓展。

用工程化视角通俗解读本次合作:无问芯穹聚焦供给端,持续压降Token生产成本、提升生产效率;MiniMax聚焦需求端,依托海量场景放大Token商业价值、扩容市场体量。看似是上下游抱团共生,实则释放了更深层的行业信号:AI“模型超市”已从行业概念,正式迭代为全产业通用标准。

算力交易中心:搭建Token经济的标准化批发市场

Token,已然成为AI时代最具通用性的数字“硬通货”。2026年3月,国内Token日均调用量突破140万亿次,相较2024年涨幅超1000倍,Token经济迎来爆发式增长。

但行业长期存在结构性短板:定价权高度集中于头部模型厂商,开发者与企业只能被动接受定价。同款Token在不同厂商、不同模型、不同场景下,价格差距高达上千倍,杂乱无序的定价体系,印证了行业尚未形成成熟、统一的交易规则。

无问芯穹联合创始人兼CEO夏立雪提出,要将行业碎片化、异构化的分散算力,转化为低成本、高性能、易调用的标准化Token,让用户无需关注底层芯片与硬件差异,只需为有效Token消耗付费,彻底简化AI使用门槛。

无问芯穹的核心布局,就是为AI时代的“新石油”Token搭建一座专业交易枢纽,将传统简单的商品买卖式Token交易,升级为标准化、规范化的金融级清算结算体系,四大核心机制彻底重构行业交易规则:

  • 实时清算结算:供应商按实际Token供货量同步回款,资金即时到账,彻底解决行业回款周期长、现金流承压的痛点,零财务周转压力。

  • 早期入驻零佣金:平台以短期利润让利换取产业流动性,早期入驻供应商全程零抽佣。入驻商家越多、Token品类越丰富,平台生态网络效应越强,形成良性产业循环。

  • API市场化竞价:打破厂商单方面垄断定价,API价格随市场供需动态浮动。需求低谷降价释放中小开发者长尾需求,需求高峰涨价激励供应商扩容增产,实现供需双向共赢。

  • 全链路生态基建:搭建涵盖商家入驻、Token交易、资金清算、数据统计的完整系统,供应商无需投入额外运营成本,只需专注核心的Token生产与供给。

Token工厂:从“卖硬件算力”到“卖有效产出”的产业跃迁

在2026世界人工智能大会(WAIC)上,无问芯穹正式发布“前店后厂一中心”核心战略,彻底颠覆传统算力租赁模式,实现Token规模化、低成本量产:

  • 一中心·算力集散枢纽:全域纳管全国碎片化异构算力资源,智能调度适配各类场景。目前已整合37000P优质算力,兼容16类主流芯片,最大化盘活闲置算力资源。

  • 后厂·Token智能生产工厂:以芯片算力、电力能耗为基础生产资料,依托全栈自研软件优化技术打造专属生产体系,高效量产标准化Token。过去一年,平台将整体推理成本压缩至原先十分之一;通过跨集群异构PD分离架构(PDD),实现首Token延迟降低51.5%、单Token成本下降37.5%。夏立雪透露,平台单日Token调用量自去年12月至今年7月暴涨40倍,未来推理成本仍有90%的优化空间。

  • 前店·AI生产力服务商店:将标准化Token产出进行场景化包装,转化为各行各业可落地、可复用、高适配的AI生产力服务,打通技术到商业的最后一公里。

三层架构形成闭环自驱飞轮:前端市场需求反向优化后端Token生产效率,进而调动算力中心全域资源,持续迭代升级产业服务能力。

模型超市全面普及:百模一键调用成为行业标配

算力交易中心解决Token流通定价问题,Token工厂解决Token生产成本问题,而遍地开花的“模型超市”,彻底解决企业模型选型、适配、运维的效率难题,无需为不同模型单独适配、充值、对账。

2026年以来,全国各地Token运营中心、模型超市密集落地,标准化AI服务体系加速成型:

长三角(嘉兴)Token运营中心:7月30日正式启动上线,企业一次接入即可按需调用DeepSeek、通义千问等百余款主流大模型。秉持“企业定任务、平台选模型”的智能理念,自动匹配当前场景成本最优模型,同时实现五大统一能力:统一API入口、统一Token计量、统一费用结算、统一政策抵扣、统一安全审计。

杭州模型超市2.0:依托阿里云百炼搭建,单密钥即可调用通义千问、DeepSeek、智谱、Kimi、MiniMax、火山豆包等数百款模型,一次接入全域复用,帮助企业降低70%以上的研发对接工作量。

中国移动MoMA模型聚合平台:汇聚300余款开源、闭源、自研大模型;无锡市级“词元超市”上线即入驻20余款企业主流模型,实现全品类模型一键调用。

毋庸置疑,数百模型一键调用,已从单一平台的特色功能,升级为全国AI产业的通用标配。

模型超市普及,重塑开发者与企业使用逻辑

模型超市的全面落地,从根本上改写了企业AI应用的开发、选型与成本管控逻辑,带来三大核心变革:

1. 接入成本近乎归零:传统模式下,接入一款模型需研读文档、编写适配代码、反复调试排错,换模型即重开发。如今一次接入、一套API、一个密钥,即可自由切换百款模型,研发工作量削减70%,业务落地周期从数周压缩至数天。

2. 模型选择权回归用户:语义精度不足可随时切换模型,DeepSeek适配性差可换Kimi,场景不匹配可换MiniMax,模型切换从“数天开发工作量”简化为“单次API参数修改”。平台智能匹配功能如同专属AI买手,自动权衡成本与效果,筛选最优方案。

3. 成本透明可控、精细化管理:模型超市2.0搭载Token比价、成本优化、数据可视化能力,调用体量、费用明细、使用趋势一目了然。企业无需多平台对账充值,单一账户即可完成全模型成本统一管控。

工程师视角:产业底层逻辑的彻底重构

无问芯穹与MiniMax的战略合作,看似是企业间的商业联动,放在全国算力基建、Token工厂、模型超市规模化落地的大背景下,其产业变革意义尤为深远。

算力交易中心规范Token流通定价、Token工厂优化Token生产成本、模型超市简化Token应用落地,三大体系深度联动,构建起Token从生产、流通到终端消费的完整产业闭环。

无问芯穹的市场化竞价机制,让Token价格彻底摆脱厂商垄断,由真实供需关系自主定价:低谷普惠开发者,高峰激励供应商,真正实现了数字代币该有的公允交易逻辑。

自8月5日签约合作起,“模型超市”彻底告别概念化标签。它是嘉兴的百模生态、杭州的一键复用、移动MoMA的三百模型定制能力,是覆盖全国的标准化AI服务基建。

模型超市时代已然全面到来,而入场门槛,仅仅是一套API、一个密钥、一次极简接入。

无问芯穹总经理曾书霖表示,团队从7人、不足50平米的初创规模,成长为300余人的专业创新团队,携手模速空间、上海仪电、沪芯沪模等上下游企业,打造全国首个算力生态超市,始终坚守“让每一度电、每一颗芯片,产出更多优质词元”的产业初心。

当AI能力如同水电般普惠透明、随取随用,模型超市便不再是营销口号,而是产业智能化转型的全新起点。无问芯穹Agentic Infra“前店后厂一中心”战略、MiniMax海量应用流量、各地Token运营中心的标准化服务能力三者深度耦合,推动模型超市完成从概念到行业标准的完美跃迁,重新定义全球Token经济的底层基础设施。

Unexhub平台核心权益

Unexhub重构全球Token交易逻辑,搭建顶尖的Token聚合分发与算力交易中心。平台汇聚Gemini、ChatGPT、Claude、DeepSeek、GLM等海内外头部大模型,同时吸纳个人、中小企业优质定制模型,依托海量供应商竞价机制与毫秒级智能路由,打造极致性价比AI服务。

开发者、企业用户按Token实际消耗量付费,最低可享官方市场价3折,高强度批量调用无惧成本压力;AI供应商可一键快速入驻,调用分成实时结算,高效变现、无资金周转压力。