iFlytek Gives Away Billions of Tokens: Grab Qwen Models at a Deep Discount

AI has exploded over the past two years, and many companies have handed parts of their business directly to AI, such as customer service, knowledge bases, and development.

To be fair, business efficiency does go up, but the Token bill at the end of the month is genuinely scary.

How to bring AI usage costs down is now an urgent problem for many companies and teams.

Last month we walked everyone through grabbing free unlimited Qwen model usage on the iFlytek Xingchen MaaS platform. Recently a new campaign appeared: the Qwen3.6-35B-A3B and Qwen3.5-35B-A3B models are directly 40% off.

If your average daily model calls reach 3 billion Tokens, you get a voucher worth 30% of the actual weekly payment returned.

If the daily average reaches 10 billion, the full amount is returned, which is essentially free, another round of savings.

It is a bit like the tiered pricing of utilities in life, except reversed. With water and electricity the more you use, the higher the unit price. This platform’s campaign returns more the more you use, clearly aimed at enterprises and high-volume teams.

More importantly, the returned vouchers are not tied to a specific model; all models in the platform’s model marketplace can be used.

It includes dozens of popular mainstream models such as iFlytek Spark, DeepSeek, Qwen, Kimi, and GLM.

However, to get the rebate you must first add the platform’s technical support on WeChat, and a technician completes the configuration. The specific process is explained below.

So how do you grab this deal? Let us walk through it.

First open the iFlytek Xingchen MaaS platform: maas.xfyun.cn/modelSquare

Enter the model marketplace and find the Qwen3.6-35B-A3B or Qwen3.5-35B-A3B model:

Then click “API Call,” fill in any model service API name, just follow the system’s recommendation.

If there is no app to authorize, first click “Create App,” go to the iFlytek Open Platform, create the app, and come back.

After clicking “Confirm,” the API call application is done and you enter the model service page.

Here, the modelId and APIKey you get support both OpenAI and Anthropic protocols for calling the API.

If we previously integrated with these two vendors’ interfaces, you only need to change the address and swap the Key to switch over.

For individuals or teams, you can also connect it to the Agent tools you use, such as WorkBuddy:

There are three steps to claim the rebate.

First return to the Qwen3.6-35B-A3B model detail page, hover over “More Discount Consultation,” and add the staff member.

Then tell them your average daily call volume and business scenario to confirm which rebate tier you qualify for.

• Average daily usage reaching 3 billion Tokens: a weekly voucher worth 30% of the actual payment.
• Average daily usage reaching 10 billion Tokens: a weekly voucher worth the full actual payment.

After the staff completes the configuration, settlement can proceed, and the returned voucher is credited directly to your account.

Let us talk about what kind of business volume in a team touches the 3-billion-Tokens-per-day threshold.

I took stock and found several common usage scenarios.

First is intelligent AI customer service. A single inquiry often spans a dozen conversation rounds, each carrying historical context. Add the fact that the AI retrieves from a knowledge base, and for products with large consultation volume you can imagine how terrifying the Token consumption is.

Another is internal AI assistants. Many companies now hand daily tasks like daily reports, weekly reports, and attendance to AI. Although these look simple, if the company has many people, the Tokens consumed add up.

There is also AI-assisted document analysis, such as contracts, financial reports, and bids that run to hundreds of pages.

These scenarios share one thing: they do not demand especially high model capability, but they do consume a lot of Tokens.

The Qwen3.6-35B-A3B model fits well, I think. It is a mixture-of-experts multimodal model, fast to respond and low cost.

That said, for complex business scenarios that need deep reasoning, you should still use a flagship model when needed.

But the bulk of a company’s daily Token burn is exactly those frequent everyday calls that do not need a top-tier model. If you move this part to cheaper models, I believe the model bill will drop visibly.

For enterprises or teams with large call volumes, this deal is worth grabbing.

Address: maas.xfyun.cn/modelSquare