TokenDance

• 1. overview (787 chars)

  TokenDance is a one-stop large model API calling platform led by Guanyun Platform, jointly developed by the Agent Universe team, and provided with underlying technical support by Wuwen Xinqiong. It is called the "Chinese version of OpenRouter" and provides a unified model API gateway for independent developers, entrepreneurial teams and enterprises with its four core advantages of "comprehensive, fast, stable and cost-effective" - ​​users only need an API Key to call domestic mainstream large models such as DeepSeek, Tongyi Qianwen, GLM, MiniMax and Kimi. It is compatible with multiple protocols such as OpenAI, Claude and Gemini, and can be accessed with zero migration without modifying the code. The product was launched as a beta version in July 2026. During the internal testing period, a tens of billions of Token subsidy program was simultaneously launched to reduce the model calling threshold for AI applications.

• 2. Product background (1647 chars)

  The birth of TokenDance directly responded to a significant pain point in the domestic AI developer ecosystem. As the number of large models explodes, various types of models such as dialogue, grammatical diagrams, audio and video generation, and intelligent retrieval are in full bloom. However, if developers want to integrate multiple models into their products, they need to register accounts with various cloud vendors one by one, manage multiple sets of API Keys, and adapt to different protocol interfaces. At the same time, reconciliation is complex and cost control is difficult. For individual developers and early entrepreneurial teams, the cost of model calling itself is a considerable burden, and the operation and maintenance costs of connecting multiple manufacturers make it even worse.   Guanyun is an AI product community that connects developers and real users. It has launched more than 2,000 AI products in less than a year, attracting tens of thousands of "watchers" and hundreds of thousands of active users. A large number of independent developers, one-person companies, and start-up teams in the community ecosystem are building in Public, and they have an extremely urgent need for a unified entrance for model calling. The Guanyun team joined forces with Agent Universe (a well-known AI agent and tool ecological team) and introduced Wuwen Xinqiong as a strategic technology partner to jointly launch the TokenDance platform.   Wuwen Xinqiong is an enterprise-level MaaS platform service provider with software and hardware collaboration and multiple heterogeneous technologies as its core. Its role is positioned as the "Token factory" behind TokenDance - responsible for efficiently converting computing resources into callable model tokens, providing the platform with full-stack enterprise-level system capabilities from model access, traffic management, inference optimization to resource scheduling.

• 3. Product functions and usage experience (2432 chars)

  The core positioning of TokenDance is "unified model API gateway". From a functional perspective, it does not solve the problem of "what to do" but "how to adjust" - so that developers do not need to expend energy on the model supply side and focus back on product innovation itself.   Multi-protocol compatibility is the most direct value point of TokenDance. It natively supports three mainstream text protocols: OpenAI, Claude, and Gemini, and covers multi-modal capabilities such as image generation, video generation, and text-to-speech. Developers only need to change the Base URL in the existing code, fill in the TokenDance gateway address, and replace the API Key to access the domestic head model without modifying any business logic.   Intelligent routing automatically matches the optimal supplier endpoint based on the model name. Developers only need to remember the unified model ID and do not need to care about which cloud vendor is behind the model, whether there is an updated version, or whether the vendor has been switched. The system automatically completes route distribution at the bottom layer, shielding the complexity caused by supplier switching.   The fault-tolerant degradation mechanism is the core design for production environments. Each model can be configured with multiple supplier endpoints and automatically switch when the primary endpoint is unavailable; a single request can also specify multiple candidate models and downgrade them step by step according to priority. This means that even if a model service used to support production fails, the business will not be interrupted.   Unified billing solves the confusing situation of recharge reconciliation on multiple platforms. Token consumption generated by all model calls is uniformly included in the TokenDance account balance. One bill covers all suppliers, making management and accounting costs clearer.   Out of the box, TokenDance supports one-click login with a Watcha account. From registration to obtaining the API Key and making the first request, it can be completed in a few minutes. The platform is compatible with the existing OpenAI SDK, and existing Python, Node.js, and cURL codes can run through just two lines of configuration changes.   The smart load balancing feature deserves a mention. TokenDance will intelligently allocate request traffic among multiple suppliers based on indicators such as price, latency, and throughput, and automatically select the current optimal supplier endpoint to ensure that developers can get the fastest response at the lowest cost at every point in time.   The platform also provides complete API call logs and usage statistics functions, and supports viewing call details and token consumption by dimensions such as model, supplier, and API Key, making it easier for developers to conduct refined cost analysis and optimization.

• 4. Pricing and business model (1534 chars)

  TokenDance’s pricing strategy adopts a two-tier structure of “transparent pricing + subsidy plan”. The platform prices each model independently, and all prices are open and transparent on the model list page. Developers can clearly know the unit price of each model before calling.   More importantly, TokenDance launched a large-scale "Ten Billion Token Subsidy Plan" during the internal beta period, open to applications for independent developers, entrepreneurial teams and individual creators. Based on the applicant's product stage and usage needs, the platform will evaluate and sponsor the corresponding Token quota to help early projects reduce the financial pressure of model invocation.   In addition, TokenDance also cooperated with Shanghai Pudong Development Bank to launch a "Pu'erxuan" co-branded physical card. The first 200 users can receive nearly 10 million Tokens (equivalent to about 100 yuan). The platform also has a "Guanxi Developer Program". After joining, you can get valuable token quota, technical support and the opportunity to experience new models first.   From a business model perspective, TokenDance itself plays the role of aggregation and distribution, and its core value lies in reducing the frictional costs of multi-vendor management for developers. The main sources of revenue are the routing price difference of model calls (the difference between the supplier's wholesale price and the platform's pricing), and possible future enterprise-level value-added services (such as privatized deployment, customized routing strategies, audit logs, etc.). The essence of the subsidy program is a customer acquisition method - attracting developers by lowering the trial threshold, and converting developers into long-term paying users after they form dependence.

• 5. User reviews (1244 chars)

  TokenDance is currently in the Beta stage, and user feedback is generally positive, mainly focusing on the two dimensions of worry-free and money-saving.   In terms of positive reviews, developers generally recognize the convenience of "one key to connect all models", especially those users who have switched from OpenRouter. They believe that TokenDance's localization is better - full coverage of domestic models, better latency than direct connection, and no need to worry about network problems. Start-up teams and one-person companies especially like the subsidy program. “It’s like trying out several models for free before deciding which one to use” is a common feedback. Unified billing and management panels are also considered features that “save money but also save time.”   Negative feedback mainly focuses on the number of models and service stability. Since the Beta phase is still in the expansion period of model access, some users reported that "the models they want to use are not available yet" and "some less popular open source models cannot be found." A few users have encountered request timeout or slow response problems, which are speculated to be related to the load strategy of the underlying provider's endpoints that is still being tuned. In addition, due to the reliance on the Watcha account system, some non-watch users feel that there is “one more step in the registration process” and hope to support more third-party login methods in the future.

• 6. Industry evaluation (1660 chars)

  TokenDance is widely referred to as the "Chinese version of OpenRouter" by industry media. This analogy itself illustrates the market's expectations for it. OpenRouter has verified the business model of unified model gateway overseas - it allows developers to flexibly switch and combine different models to complete different tasks without having to bind to a certain model supplier. TokenDance is doing exactly this in the domestic market.   As the underlying technical supporter, Wuwen Xinqiong has injected enterprise-level model reasoning and optimization capabilities into TokenDance. According to official data, its precision alignment rate exceeds 99.9%, throughput is increased by 2 to 3 times, overall latency is reduced by 50%, and first-word latency is within 500 milliseconds. If these data are true and reliable, it means that TokenDance is even better than some cloud vendors that developers can directly connect to in terms of performance.   From the perspective of the competitive product landscape, there is currently no truly “unified model gateway” head product in the domestic market. Each cloud vendor has its own model API service, but no platform is willing to actively aggregate competitors' models. TokenDance's third-party neutral positioning is precisely its differentiating advantage - it does not require taking sides and allows users to freely choose which model to use behind each task. Users can also switch at any time without any binding.   The potential risk is that once a leading cloud vendor launches similar aggregation services (such as Alibaba Cloud integrating GLM and DeepSeek at the same time), TokenDance's neutral advantage will be weakened. However, judging from the current market situation, various cloud vendors are more inclined to lock users within their own ecosystems and have insufficient motivation to launch aggregation services. TokenDance will have a window period at least in the next one to two years.

• 7. Controversies and Risks (1829 chars)

  The first is stability and SLAs. As an aggregation platform, TokenDance's service availability is highly dependent on the stability of the underlying providers. Although the platform is designed with a multi-supplier fault-tolerant downgrade mechanism, if multiple suppliers fluctuate at the same time (for example, a popular model is restricted due to a surge in users), TokenDance's upper-level experience will also be affected. There is currently no public SLA commitment on the platform. For enterprise users who use TokenDance for production environments, this is a risk point that needs to be evaluated.   Second is data security and privacy. All request content forwarded through TokenDance will pass through its middle layer, and developers need to confirm whether the data in the request involves sensitive information. The platform has instructions on data transmission encryption and access control, but there is currently a lack of detailed information disclosure on whether data from industries with strict compliance requirements such as medical care and finance meet the corresponding data protection regulations.   The third is the sustainability of the business model. Subsidy programs are an effective means of acquiring customers, but in the long run, platforms need to make users willing to pay for convenience after the subsidies end. The current core value is "saving trouble", and how much premium "saving trouble" can support still needs to be verified among the domestic developer community. If most developers only use TokenDance as a transition tool for the trial model, and then directly switch to supplier direct connection after confirming the selection, the platform's long-term user retention will face challenges.   Finally, there are the boundaries of model coverage. Currently, the main ones that have been connected are domestic head models. For the fast iteration models of the open source community (such as the small special models that are frequently updated on Hugging Face), the access rhythm still needs to be followed up. If users’ model needs exceed TokenDance’s coverage, the platform’s appeal will be compromised.

• 8. Applicable groups and usage suggestions (1030 chars)

  TokenDance is best suited for two types of users. The first category is individual developers and start-up teams who are doing product prototype verification. They need to quickly trial and error multiple models to find the optimal solution. TokenDance's zero-migration access and subsidy plan can significantly reduce the cost of trial and error. The second category is teams that use multiple models in a production environment. Unified management and automatic downgrade mechanisms can simplify operation and maintenance work, so that teams do not have to maintain a separate set of call links for each model.   For users in industries that have strict requirements for data privacy (finance, medical, government affairs), it is recommended to confirm whether TokenDance meets the data compliance requirements of their industry before use, and consider deploying a privatization plan if necessary (if launched in the future).   For users who only use a single model, direct connection to the manufacturer may be a more economical choice - after all, the added value of TokenDance lies in "managing multiple models." If you only use one model from one company, going around a layer of gateways will add unnecessary delays.

• 9. Summarize (583 chars)

  TokenDance solves a real and specific pain point: the threshold for domestic developers to access large models is too high. It uses an API Key, a billing system and a gateway to integrate the mainstream AI models in the market, allowing developers to focus on the product itself. Backed by the developer resources of Guanyun Ecosystem and the underlying computing power support of Wuwenxinqiong, TokenDance has a good starting advantage in both technical and ecological dimensions. Whether it can become a true "Chinese version of OpenRouter" depends on the expansion speed of model coverage, the degree of stable service delivery, and how many users are willing to stay after the subsidy ends.

No in-depth report yet

• 

  Jonathan_PatelIII

  —

  I just applied for the TokenDance subsidy today and filled out a form. It was approved the next day and I was given a 5 million Token quota, which was enough for me to run experiments for a while.

---

• 

  SamuelFoster_2020_241

  —

  After running with TokenDance for a day, changing the Base URL is really simple. It turns out that OpenAI’s SDK can be used directly, and it can be done with just a few lines of code.

---

• 

  JudyJohnson_Pro97

  —

  It's so cool to be able to connect all models with one key. Before, I had to register for five platforms, and I had to find the corresponding key every time I changed models. Now I finally don't have to worry about these messy things.

---

• 

  飞鸟_12

  —

  There are still a few models. Several open source small models I want to use are not available. I hope to expand them soon.

---

• 

  Andrew.Turner5200

  —

  Switched from OpenRouter, the latency is lower than direct connection. The key is that there are all domestic models, so there is no need to worry about network problems.

---

• 

  RavenRocketRodriguez

  —

  The subsidy is so good that the student party directly saves a lot of money when doing graduation projects. The 20 yuan quota is enough to last for a while.

---

• 

  WhaleAlertHall

  —

  When I first started using TokenDance, I reported an error and automatically switched to the alternative model, only to find that it automatically downgraded. This design is very stable.

---

• 

  VWright_2023

  —

  I hope to support more login methods. Currently I can only use Watcha to log in, which is sometimes inconvenient.

---

• 

  Bobby.WalkerQ96

  —

  I recommended TokenDance to a friend who works as an agent. He said that unified management of multiple model calls just solves their current pain points. In the multi-agent system built by their team, each Agent may adjust different models. Previously, a bunch of API Keys and billing rules had to be maintained. After changing to TokenDance, unified management and automatic downgrade directly solved two long-standing problems.

---

• 

  KatherineMorris_Pro107

  —

  After using it for three days, I feel good overall. Several mainstream models can be used, and all usage can be viewed in one backend, making management much easier.

---

• 

  SOall

  —

  I encountered a request timeout in the afternoon. It may be that the load of the new platform is still being adjusted, but I just switched the model and it was fine. It was not a big problem. But what I tested was the GLM model. After switching to Tongyi Qianwen, the response speed was significantly faster, which shows that there are still gaps between suppliers. Intelligent routing itself is fine, but the supplier capabilities themselves may be uneven.

---

• 

  Kelly_Gutierrez_8806

  —

  The unified billing function is so important. In the past, monthly reconciliation would drive people crazy. Now, you can see all the bills in one place, which saves so much time.

---

• 

  安然991

  —

  The idea of ​​TokenDance should have been developed a long time ago. There are more and more domestic models, each of which needs to be connected separately, and developers are not specialized in integration.

---

• 

  Richard.King

  —

  I applied for a tens of billions subsidy, filled in the product stage and usage information, and was given 8 million Tokens, which is enough for our team to use for half a year.

---

• 

  BlockVenturesRamirez

  —

  To be honest, I was a little worried about stability when I first launched it. After running it for more than a week, there were basically no problems and it was better than expected.

---

• 

  MorganChevalier

  —

  It saves our one-person company a lot of trouble because we don’t have to charge money on every platform. Moreover, intelligent routing selects the best supplier, and the price and speed are pretty good.

---

• 

  Victoria_Coleman_7058

  —

  There is a small problem. The model ID of some sample codes in the document does not seem to match the actual one. I hope the official updates it soon.

---

• 

  DeFiDex_eth

  —

  It is much more convenient than directly connecting to Silicon Mobile, Alibaba Cloud, etc. You don’t need to remember a bunch of API addresses, just one entrance.

---

• 

  GraceHernandez

  —

  The co-branded card activity is also very good. When you apply for the Pu'erxi card, you will receive 10 million tokens, which is equivalent to 100 yuan for free.

---

• 

  云烟_17

  —

  I use it in automated scripts and run tens of thousands of requests every day. There has been no drop in the chain so far. Fault tolerance and downgrading really work.

---

• 

  ydd50

  —

  I am working on a SaaS product for AI customer service. Previously, each customer had to configure the model supplier separately, and the access process was extremely long. After using TokenDance, one API Key can be used to handle all models. Whether customers want to use DeepSeek, GLM, or Tongyi Qianwen, the back-end code does not need to change a single line, and the docking cycle is shortened from two weeks to three days. Moreover, the subsidy plan approved 30 million tokens for us, and the reasoning costs in the first few months were basically zero burden, which is too critical for the start-up team.

---

• 

  MateoThomas

  —

  For those who make AI products, I highly recommend getting one. Use the time saved to work on product logic, and don’t waste time on model access. Previously, three members of our team spent two whole weeks to connect to the APIs of five model platforms, write adaptation codes, and conduct billing reconciliations. After all the hard work, everyone was numb. Switching to TokenDance, it can be done in one afternoon, and the remaining time is spent on product optimization. The efficiency improvement is too obvious.

---

• 

  Harold_Bennett520

  —

  After a comparison, there is indeed no similar product in China that can achieve this level of integration. OpenRouter is slow to use in China, and TokenDance can fill the gap.

---

• 

  RLee_Pro0

  —

  I applied for closed beta when it was in Alpha and used it all the way to Beta. Let me talk about my experience of using it. The biggest advantage is that you can switch models without changing the code. I use OpenAI's Python SDK for the backend. I can directly change the Base URL and API Key. The compatibility is much better than expected. The experience of smart routing is also good. The same model automatically switches between different suppliers, and I basically don’t feel any interruption. Disadvantages: First, some unpopular models are indeed not covered. After checking, there are currently only 8 suppliers listed, and iFlytek is not yet available. Second, the billing page sometimes loads a bit slowly, and the query efficiency of API call logs needs to be improved when the amount of data is large. Overall, the beta stage has been completed very well, and we look forward to the official version.

---

• 

  Sean_SandersK

  —

  Intelligent load balancing is indeed useful. I ran several rounds of comparison tests and found that routing the same request using TokenDance was almost 15% cheaper than manually selecting a provider.

No reviews yet