AI Card Bao GGOPHDEVELOPER.AICARDBAO.COM

Industry News

The bean bag work reduced the magnification of the two models to 0.01, and one 3D modeling cost about 23 cents.

Doubao has adjusted its billing rules. The rate at which quotas are consumed for several models has been reduced; in particular, the reduction amount for DeepSeek V4.1 Flash and Doubao 2.1 Lite has been set to 0.01.

What does a multiplier of 0.01 mean? For the same task, previously 100 points would be deducted from the account; now only 1 point is deducted. For simple conversations and tasks that require minimal resources, no fees will be charged at all.

The official provided several actual measurement prices. The cost of performing 3D modeling is approximately 0.23 yuan per time, while creating a dynamic web page costs around 0.06 yuan per task. According to their statistics, more than 80% of daily office tasks can be completed for free using these services.

This “80%” actually depends on how the denominator is calculated. In daily office work, the majority of tasks involve simple activities such as rewriting, summarizing information, or searching for data – it’s no surprise that these account for 80% of the total workload. The real challenge comes from the remaining 20%, which includes tasks like analyzing long documents, handling multi-step intelligent system operations, or carrying out tasks that require code execution. For these tasks, models with a scaling factor of 0.01 are usually not sufficient.

So, the actual experience is roughly like this: Usually, using it casually doesn’t cost anything at all; it’s only when you occasionally run a heavy-duty task that you’ll notice your usage quota decreasing.

Also launched simultaneously were features such as signing in to earn credits and a 30-day trial experience for new users – a typical combination of strategies designed to attract new customers. The list of available models now includes third-party solutions like Kimi K3.1 Pro and GLM 5.3. On ByteDance’s platform, users can now choose models from both MoYi AnMian and Zhipu. This would have been virtually unimaginable just a year ago.

The logic behind this approach is quite clear. In the category of office intelligent agents, companies like Tencent, Alibaba, and ByteDance are all competing fiercely. The functional differences between these products are becoming increasingly minimal; what truly distinguishes one product from another is whether users feel comfortable using it. Only by reducing the marginal cost of such lightweight tasks to zero can users develop the habit of delegating everything to these intelligent agents.

The fact that such inexpensive models are available at such low prices also indicates that the computing costs associated with using Flash for inference have dropped to the point where they can even be considered as part of the cost incurred in acquiring new customers.

Related