
When Model Routing Meets Compute Trading: The Token Economy Is Being Repriced
English First|中文版本见后
On July 23, 2026, Runway introduced what might be called a “quiet disruption”: Media Router, an intelligent model-routing system built specifically for generative media.
Around the same time, another project—Unexhub—began drawing attention with a broader proposition: reshape the global distribution of AI Tokens by building a compute trading center that supports API bidding, real-time settlement, and zero-commission incentives for early suppliers.
Viewed separately, these two developments appear to serve different parts of the AI stack.
Viewed together, however, they reveal the same underlying shift.
Media Router helps developers reduce costs when consuming AI services. Unexhub aims to lower costs when sourcing and distributing AI capacity.
One optimizes spending on the demand side. The other attempts to introduce market-based pricing on the supply side.
As model routing and compute trading begin to converge, the Token economy is moving beyond simple API pricing toward a more dynamic system built around timing, task requirements, supply, demand, and execution efficiency.
Media Router: Not Just Choosing a Model, but Spending Tokens More Intelligently
The basic logic behind Media Router is straightforward, but highly practical.
Instead of requiring developers to compare models manually, the system evaluates each request across three main dimensions:
Output quality
Processing speed
Cost
The routing system then selects a model according to the priority of the specific task.
Developers no longer need to repeatedly ask:
Which model is powerful enough?
Which model is fast enough?
Which model offers the best price-performance ratio?
The router turns those decisions into an automated process.
Runway broadly divides generative media workloads into three categories.
High-Speed Generation
During ideation and creative validation, speed matters more than final quality.
The goal is to produce drafts quickly, test visual directions, and determine whether an idea is worth developing further.
Slower, Higher-Quality Generation
During final production, image quality, consistency, detail, and precision become more important than speed.
This is where higher-end models can be selected for final delivery.
Balanced Quality at Lower Cost
For routine content operations and large-volume production, teams often need acceptable quality without paying premium-model prices for every request.
This category is particularly important for marketing departments, content studios, and platforms generating media at scale.
In practice, a lightweight model may be used to create early video drafts, while a more advanced model is activated only during the final rendering stage.
Through a single Runway Dev integration, developers can access models and services from multiple providers, including Seedance, GPT Image 2, ElevenLabs, and others.
The key value is not merely convenience.
It is the ability to match the cost of the model with the value of the task.
A quick concept test should not consume the same budget as a final commercial asset.
Token Costs Are Becoming an Operating Issue
A particularly revealing point in TechCrunch’s coverage was that Token pricing had become one of the most discussed issues of 2026, especially among companies investing heavily in intelligent AI applications.
The reason is simple.
As enterprises move from occasional chatbot use to continuous AI workflows, the cost structure changes completely.
AI systems are now being used to:
Generate and edit media
Run customer service agents
Analyze internal documents
Write and test code
Process audio and video
Execute multi-step business tasks
A single request may be inexpensive, but millions of requests, repeated generations, failed outputs, and high-resolution production can quickly create substantial bills.
Runway had previously shifted its unlimited subscription offering toward Token-based billing. The later introduction of Media Router may seem contradictory at first: the platform charges by Token, yet also helps customers reduce Token consumption.
In reality, the two strategies reinforce each other.
When customers can see that the platform actively helps control costs, they are more likely to move larger workloads onto it.
The platform is not simply selling more Tokens.
It is selling a more efficient way to use them.
Unexhub: From Selling Compute Capacity to Trading Tokens
While Media Router optimizes the demand side, Unexhub is attempting to restructure the supply side.
In engineering terms, Unexhub’s objective is to turn Tokens from isolated resources controlled by individual model providers into resources that can be aggregated, routed, distributed, and priced through a shared market.
Its proposed compute trading center can be understood through three core mechanisms.
Layer One: Token Aggregation and Distribution
AI capacity is highly fragmented.
Large model providers control major infrastructure, while smaller enterprises, independent providers, regional compute centers, and individual developers may also possess valuable models or available inference capacity.
The problem for smaller suppliers is often not the quality of their resources.
The problem is distribution.
They may have available compute but lack customers. They may offer competitive models but lack traffic, billing infrastructure, routing technology, and bargaining power.
Unexhub aims to aggregate suppliers of different sizes into a unified distribution network.
Through standardized API access, centralized scheduling, and unified metering, available capacity can be distributed to enterprise users and developers without requiring each supplier to build a complete sales and settlement system independently.
For suppliers, this creates an additional commercialization channel.
For developers, it creates a broader pool of models, endpoints, and price options.
Layer Two: Real-Time Settlement and Lower Cash-Flow Pressure
Traditional enterprise settlement often involves long billing cycles.
A supplier may deliver computing resources today but wait weeks or even months for payment.
For smaller providers, delayed settlement can become a more serious constraint than technical capacity itself.
Unexhub proposes a real-time settlement model in which revenue is calculated and distributed as API calls are completed.
In other words:
When capacity is consumed, the corresponding revenue can be recognized and settled more quickly.
This mechanism can shorten payment cycles and reduce the working-capital pressure faced by small and medium-sized suppliers.
It also changes the economics of participating in an AI distribution network.
Instead of committing resources and waiting through lengthy reconciliation processes, suppliers gain a clearer view of usage, revenue, and settlement performance.
Early suppliers may also qualify for promotional commission rates as low as 0%, lowering the cost of testing the platform and building an initial customer base.
The value of this incentive is not simply “free access.”
It gives suppliers time to validate demand, observe usage patterns, and establish distribution channels before taking on a standard platform commission structure.
Layer Three: API Bidding and Dynamic Market Pricing
The most ambitious part of the Unexhub model is API bidding.
Under conventional pricing, an AI provider publishes a fixed price, and users either accept it or choose a different provider.
A trading mechanism introduces a different logic.
API prices can respond to:
Real-time supply
User demand
Model performance
Response speed
Available capacity
Peak and off-peak periods
Supplier competition
During periods of lower demand, developers may access available capacity at lower prices.
During peak demand, suppliers with stable capacity and reliable performance may receive higher returns.
This creates benefits for both sides of the market.
Developers gain more opportunities to reduce costs, while suppliers gain a mechanism for monetizing idle capacity and earning more when demand rises.
The price is no longer determined exclusively by the provider’s official rate.
It becomes a market signal.
Why Early Zero-Commission Incentives Matter
A marketplace is only useful when it has sufficient supply.
Without enough suppliers, there is no meaningful price competition, routing flexibility, or redundancy.
This is why Unexhub’s early-stage zero-commission policy is strategically important.
The platform is effectively reducing the cost of joining during the supply-building phase.
Suppliers can connect models and computing resources, receive traffic, observe demand, and settle usage revenue without immediately sacrificing a portion of each transaction.
The objective is clear:
First, expand the supply pool. Then, improve price discovery, routing efficiency, and transaction volume through market competition.
For individual developers and smaller AI companies, this can lower the barrier to entering the API market.
For larger suppliers, it provides another distribution channel without requiring a complete rebuild of existing infrastructure.
One Problem, Two Different Solutions
Media Router and Unexhub address two sides of the same cost problem.
Media Router answers:
How can a developer use the right model for each task?
Unexhub answers:
How can that model capacity be sourced at a more competitive price?
Media Router reduces unnecessary spending by routing requests to the most suitable model.
Unexhub seeks to reduce procurement costs by allowing supply, demand, and pricing to interact dynamically.
One focuses on how Tokens are consumed.
The other focuses on how Tokens are supplied and priced.
Together, they point toward a new cost architecture for AI.
A developer may eventually be able to submit a task while the system automatically determines:
Which model is suitable
Which provider has available capacity
Which endpoint offers the best price
Whether the task can wait for an off-peak period
Whether a faster or higher-quality route is worth the additional cost
How much should be paid per Token for that specific request
At that point, using AI will no longer be a static purchasing decision.
It will become a real-time optimization process.
Unexhub’s Position in the Emerging Token Market
Unexhub is positioning itself as a Token aggregation and distribution center that connects AI suppliers, enterprise users, and developers through a unified compute trading network.
The platform is designed to support mainstream domestic and international AI ecosystems, including access to models and services associated with Gemini, ChatGPT, Claude, DeepSeek, GLM, and other providers.
It also aims to include high-quality models and computing resources contributed by independent developers, small teams, and medium-sized AI companies.
Instead of forcing users to maintain separate accounts, API formats, billing systems, and supplier relationships, Unexhub provides a unified access layer.
Its core capabilities include:
Aggregated access to multiple AI models
Standardized API distribution
Millisecond-level intelligent routing
Usage-based Token billing
Dynamic supplier competition
Real-time revenue settlement
Enterprise integration and customization
Flexible cost-control options
Developers can pay according to actual Token usage rather than committing to unnecessary packages.
For selected routes and models, pricing may be available at as low as approximately 30% of comparable market or official pricing, depending on supply, demand, availability, and promotional conditions.
The real value is not simply a lower headline price.
When teams are running large-scale document processing, customer service, coding, media generation, or agent workflows, even a moderate reduction in the cost per Token can translate into meaningful long-term savings.
Suppliers, meanwhile, can connect through a unified channel and receive revenue based on actual model usage, with settlement data available more quickly.
The Next Stage of the Token Economy
Runway Chief Product Officer Anthony Maggio described routing as a natural extension of the company’s ambition to become a convenient one-stop platform for accessing generative media models.
Unexhub’s ambition goes one step further.
It is not only trying to become a place where developers access models.
It is trying to become a place where Token supply itself can be aggregated, priced, routed, distributed, and settled.
When model routing and compute trading operate together, developers no longer face a simple question:
How much does this model cost?
Instead, they face a dynamic optimization problem:
Which model should be used, through which supplier, at what time, with what level of quality, and at what price?
That shift—from fixed pricing to intelligent routing and market-based allocation—may become one of the clearest signs that the Token economy is entering a more mature stage.
The future of AI infrastructure may not belong to a single model or a single provider.
It may belong to the network that can match every task with the right model, the right capacity, and the right price.
当模型路由遇上算力交易:Token 经济正在被重新定价
中文版本
2026 年 7 月 23 日,Runway 投下了一颗“温和炸弹”——Media Router,一套专门面向生成式媒体的智能模型路由系统。
几乎在同一阶段,一个名为 Unexhub 的项目也逐渐进入行业视野。它提出重塑全球 Token 分发逻辑,建设算力交易中心,引入 API 竞价、实时结算,并以早期最低零佣金的方式吸引供应方入驻。
单独来看,这两件事似乎处在 AI 产业链的不同位置。
但把它们放到同一条时间线上,就会发现二者共同指向了 Token 经济的一次深层变化。
Media Router 解决的是:AI 服务使用过程中,怎样花得更少。
Unexhub 解决的是:AI 算力采购和分发过程中,怎样买得更便宜、卖得更高效。
一个从消费侧优化预算,一个从供给侧推动市场化定价。
当模型路由与算力交易开始结合,Token 的成本结构也将不再只由模型厂商的一张官方价目表决定,而会逐渐受到任务类型、调用时间、市场供需、模型性能和路由效率的共同影响。
Media Router:不只是帮你选模型,而是帮你更聪明地使用 Token
Media Router 的底层逻辑并不复杂,却精准击中了开发者的实际痛点。
过去,开发者需要手动比较不同模型:
哪个模型生成质量更高?
哪个模型响应速度更快?
哪个模型价格更低?
当前任务是否值得调用旗舰模型?
Media Router 将这些判断归纳为三个核心维度:
输出质量
处理速度
调用成本
系统会根据单次任务的具体优先级,自动选择更适合的模型。
开发者不必每次都在多个模型之间反复切换,也不需要持续计算哪个模型更划算。
路由系统把模型选择变成了一项自动化决策。
Runway 将生成式媒体的使用场景大致分为三类。
高速生成:先验证想法
在创意构思和方案验证阶段,最重要的通常不是最终画质,而是尽快看到结果。
团队需要快速生成草稿、测试视觉方向,判断创意是否值得继续投入。
在这一阶段,系统可以优先选择速度更快、成本更低的轻量模型。
低速高质量:面向正式交付
进入成片制作、商业交付和最终渲染阶段后,画面细节、稳定性、一致性和生成精度会变得更加重要。
此时,系统可以自动切换至高精度旗舰模型。
中等质量与低成本:面向批量生产
对于日常营销素材、短视频批量生成、商品展示图和规模化内容生产而言,企业往往不需要每一次调用都使用最昂贵的模型。
他们更需要的是:
在质量可接受的前提下,把单次生成成本控制在合理范围内。
在实际工作流中,创意验证阶段可以使用轻量模型快速出样;方案确定后,再调用高质量模型完成最终制作。
开发者只需通过 Runway Dev 的统一入口,即可接入 Seedance、GPT Image 2、ElevenLabs 等多家厂商的模型与服务。
它带来的价值不只是“接入更方便”。
更重要的是,让模型成本与任务价值相匹配。
一个用于内部讨论的初稿,没有必要承担与正式广告成片相同的生成成本。
Token 成本正在从技术问题变成经营问题
TechCrunch 的相关报道中有一句话很值得关注:
Token 定价已经成为 2026 年的热门话题,全面投入智能 AI 的企业,正在真实感受到高额 Token 账单带来的压力。
原因并不难理解。
过去,企业可能只是偶尔使用聊天机器人;现在,AI 已经开始持续参与大量业务流程:
生成与编辑图片、音频和视频
运行客户服务智能体
分析企业内部文档
编写、测试和迁移代码
处理录音、会议与多媒体内容
自动执行多步骤业务任务
单次调用的价格可能并不高,但当请求量扩大到数十万甚至数百万次,再叠加重复生成、失败重试、长上下文和高分辨率输出,最终账单就会迅速增长。
此前,Runway 将部分无限订阅方案调整为 Token 计费;随后又推出 Media Router,帮助开发者降低 Token 消耗。
表面上看,这似乎有些矛盾:
平台一边按照 Token 收费,一边又帮用户节省 Token。
但从商业模式来看,这两件事恰好形成了闭环。
当客户意识到平台不仅提供模型,还能够帮助控制成本,他们反而更愿意把更多业务迁移到平台上。
平台出售的,不再只是更多 Token。
而是一种更高效地使用 Token 的方式。
Unexhub:从“出售算力”走向“交易 Token”
如果说 Media Router 是在消费侧优化,那么 Unexhub 试图推动的,则是供给侧的重构。
用工程化语言来理解,Unexhub 希望把原本分散在各个模型厂商、算力中心和开发团队手中的 Token 资源,转化为可以被统一聚合、调度、分发和定价的市场化资源。
它建设的算力交易中心,可以拆分为三层核心机制。
第一层:Token 聚合与分发
当前的 AI 算力供给非常分散。
大型模型厂商掌握着主要资源,但与此同时,大量中小企业、区域算力中心、独立开发者和个人供应方,也可能拥有性能优秀的模型或尚未充分利用的推理资源。
这些中小型供应方真正缺少的,往往不是技术能力,而是分发渠道。
他们可能有算力,却找不到稳定客户;有模型,却缺乏流量、计费系统、路由能力和市场议价权。
Unexhub 希望将不同规模的供应方聚合到统一分发网络中。
通过标准化 API 接入、统一调度和统一计量,供应方无需独立建设完整的销售、支付和结算系统,也能够把模型能力提供给企业用户与开发者。
对供应方而言,这意味着新增了一条商业化渠道。
对开发者而言,则意味着可以获得更丰富的模型、节点和价格选择。
第二层:实时结算,降低资金周转压力
传统企业服务通常存在较长账期。
供应方今天提供算力,可能需要数周甚至数月才能收到对应款项。
对于中小型模型和算力供应方而言,回款速度有时比技术能力更影响业务发展。
Unexhub 提出的实时结算机制,是按照实际 API 调用量计算收益,并更快地完成调用分成结算。
简单来说:
算力一旦被调用,对应收益就能更快确认和回收。
这种模式能够缩短账期,降低中小型供应方的现金流压力,同时让模型调用量、收入和结算情况变得更加透明。
对于早期入驻的优质供应方,平台还可提供最低 0% 佣金的阶段性政策。
这一权益的价值并不只是“暂时不收费”。
更重要的是,供应方可以在较低成本下测试市场需求、观察真实调用量、积累客户,并建立稳定分发渠道。
在正式进入常规佣金体系前,供应方已经有机会验证自己的模型是否具备商业竞争力。
第三层:API 竞价,让价格随市场变化
Unexhub 体系中最具市场想象力的机制,是 API 竞价。
传统 API 定价通常由模型厂商单方面制定。
厂商公布一个价格,开发者只能选择接受、减少使用,或者更换其他模型。
API 竞价引入了另一种逻辑。
模型调用价格可以根据以下因素动态变化:
实时供应量
市场调用需求
模型性能
响应速度
剩余算力容量
高峰与低谷时段
不同供应方之间的竞争
在需求较低的空闲时段,开发者可能以更低价格获得可用算力。
在调用高峰期,具备稳定容量和优质服务能力的供应方,则可能获得更高收益。
这套机制同时照顾了市场两端。
开发者能够降低调用成本,供应方可以盘活闲置资源,并在需求旺盛时获得更合理的回报。
API 的价格不再只是一张静态价目表。
它开始成为供需变化的市场信号。
为什么早期零佣金很重要
一个交易市场只有拥有足够多的供应方,才能真正形成竞争。
如果供应池不够大,平台就无法提供足够丰富的模型选择,也难以形成有效的价格发现、备用路由和服务冗余。
因此,Unexhub 面向早期供应方推出最低零佣金政策,其战略目的非常清晰:
降低入驻门槛,优先扩大供给池。
供应方可以把模型与算力资源接入平台,获得调用流量,观察市场需求,并按照实际用量进行收益结算,而不必一开始就承担较高的渠道成本。
这套机制的路径是:
先扩大供给,再通过竞价、路由和交易规模提升市场效率。
对于个人开发者和中小型 AI 企业而言,它降低了进入 API 商业市场的门槛。
对于大型供应方而言,它也提供了一条无需大规模改造现有系统的新增分发渠道。
同一个问题的两个答案
Media Router 与 Unexhub 解决的,本质上是同一个成本问题的两个侧面。
Media Router 回答的是:
同一个任务,应该怎样选择更合适的模型?
Unexhub 回答的是:
同一种模型能力,应该通过哪个供应方,以什么价格采购?
Media Router 通过智能路由,减少不必要的旗舰模型调用。
Unexhub 则希望通过供需竞争和动态定价,让开发者以更具竞争力的成本获得模型能力。
一个关注Token 怎样被使用。
另一个关注Token 怎样被供应和定价。
两者结合后,AI 成本管理将不再只是“选一个便宜模型”这么简单。
未来,开发者提交一个任务后,系统可能会自动判断:
哪个模型更适合当前任务
哪个供应节点仍有可用容量
哪个接口当前价格更低
任务能否等待至低谷时段执行
是否值得为更快速度或更高质量支付额外成本
当前这一次请求应该按照怎样的 Token 价格完成
到了那个阶段,AI 调用就不再是一项静态采购。
而是一项实时发生的动态优化。
Unexhub 在 Token 市场中的定位
Unexhub 正在将自身定位为连接 AI 供应方、企业用户与开发者的 Token 聚合与分发中心。
平台致力于聚合海内外主流 AI 模型生态,支持接入 Gemini、ChatGPT、Claude、DeepSeek、GLM 等模型与服务,同时也为个人开发者、中小型团队和专业模型供应方提供模型与算力分发渠道。
用户不必分别维护多套账户、API 格式、供应商关系和计费系统,而是可以通过一个统一入口完成模型调用与管理。
其核心能力包括:
多模型聚合接入
标准化 API 分发
毫秒级智能路由
按 Token 实际用量计费
供应方竞价机制
调用收益实时结算
企业级接入与定制服务
灵活的成本控制方案
开发者可以根据真实 Token 用量付费,无需购买超出需求的固定套餐。
根据模型、供应节点、市场供需和阶段性优惠的不同,部分调用线路最低可享受约为市场或官方参考价格 3 折的成本方案。
这项权益真正重要的地方,并不只是数字看起来更低。
对于需要长期运行文档处理、客服智能体、代码生成、内容生产和多步骤 Agent 的企业而言,每百万 Token 成本的下降,都会随着业务规模扩大而持续累积,最终形成可观的长期预算节省。
供应方则可以通过统一接口快速接入分发渠道,按照真实调用量获得收益,并更及时地查看和结算调用分成。
Token 经济的下一阶段
Runway 首席产品官 Anthony Maggio 曾表示,模型路由能力与 Runway 建设生成式媒体一站式入口的整体愿景高度一致。
Unexhub 的目标则更进一步。
它不仅希望成为开发者调用模型的平台,还希望成为 Token 供给被聚合、定价、路由、分发和结算的交易基础设施。
当模型路由与算力交易中心同时运转,开发者面对的问题将不再只是:
这个模型多少钱?
而会变成一道更动态的优化题:
应该在什么时间,通过哪个供应方,调用什么模型,以什么质量和速度,花费多少成本?
从固定标价走向智能路由与市场化配置,或许正是 Token 经济逐渐成熟的重要标志。
未来的 AI 基础设施竞争,未必只属于某一个最强模型或某一家最大厂商。
它更可能属于这样一个网络:
能够为每一项任务,匹配合适的模型、可用的算力,以及更合理的价格。