1. Core Judgment: AI Billing Will Shift From Token to Task, Restructuring Industry Paradigms
Xu Li pointed out clearly in his speech that as artificial intelligence enters the era of "super individuals", the underlying logic of industry pricing will undergo a fundamental shift. The mainstream token-based billing (word token consumption) will gradually give way to a pricing model centered on tasks (deliverables of complete business workflows). He stated that this shift will reshape the economic paradigm of AI and drive the inclusive popularization of AI technology.
“Everyone is competing on AI computing power and token unit prices. But in an inclusive AI era, billing should be anchored to the price of completing a full task. Shifting from tokens to tasks represents an entirely new economic calculation framework. Sustained cost reduction for standardized tasks will lower barriers to AI adoption and deliver true inclusiveness.”
2. Current Mainstream Model: Core Logic and Prominent Pain Points of Token Billing
At present, token-based billing remains the industry standard. Its core logic is to charge users based on the total number of input and output word tokens consumed during computation. This straightforward model was reasonable in the early days of AI, when users were mostly programmers and enterprise technical teams. However, as AI technology gains widespread adoption, ordinary non-technical users and small and medium-sized merchants have become major adopters, and the flaws of token billing have become increasingly prominent.
Xu Li highlighted three core drawbacks of token billing:
First,
unpredictable costs. Multi-turn dialogues and multimodal generation within AI agents cause context lengths to expand endlessly, making token consumption uncontrollable. Monthly bills for enterprises and individuals fluctuate wildly, creating major challenges for budget management.
Second,
disconnect between cost and value. Low-cost token models often deliver subpar outputs, forcing users to run repeated retries that double overall token consumption and raise comprehensive costs. Service providers only supply computing power without taking accountability for task quality or delivery results.
Third,
incompatibility with complex intelligent agents. AI agents involve retrieval, web access, tool invocation, multi-step reasoning and other complex workflows. Average users cannot parse itemized billing statements, nor can they optimize prompts to contain expenses effectively.
3. Next-Generation Model: Core Connotation and Advantages of Task-Based Billing
To address the many pain points of token billing, Xu Li introduced the innovative task-based billing framework. Its core principle is to set fixed flat fees around standardized complete business deliverables. Under this model, service providers absorb all internal computing overhead generated throughout task execution, including repeated calls, retries, multimodal processing and tool retrieval. Users only pay attention to whether the task meets delivery standards, without needing to understand underlying token consumption details.
This is analogous to ride-hailing: riders pay a fixed fare from origin to destination rather than being charged per wheel rotation. Xu Li outlined four core strengths of task billing that resolve existing industry pain points:
- Value alignment: Users pay for tangible business outcomes instead of raw computing power, with clear return-on-investment visibility.
- Stable and controllable costs: Fixed pricing for standardized tasks enables precise enterprise AI budget planning, eliminating unexpected surcharges.
- Broad compatibility: The model seamlessly supports AI agents and long multimodal workflows, abstracting complex technical details to lower user entry barriers.
- Drives industrial upgrading: To cut internal costs and boost competitiveness, vendors will optimize model performance and reduce redundant retries, continuously pushing down unit task pricing.
Internal SenseTime data shows the comprehensive cost of standard commercial tasks in Q2 2026 dropped to 1/18 of levels recorded in the same quarter of 2023, bringing AI closer to utility-grade infrastructure like water and electricity.
4. Drivers Behind the Model Shift: Underlying Transformations Shaping the AI Industry
The transition in billing models is not arbitrary; it stems from fundamental shifts reshaping the AI industry, falling into three key categories:
First, a transformation in AI user demographics. Early adopters were professional technical staff, while today’s primary users consist of the general public and small businesses. This group cannot comprehend complex token consumption metrics and prefers the simple
"pay for results" pricing logic, laying the user foundation for task billing.
Second, a flawed industry competition dynamic. The sector is trapped in homogeneous price wars centered on per-ten-thousand-token rates, with vendors overlooking output quality and total cost of ownership. Xu Li argues that industry competition should ultimately revolve around the comprehensive capability to deliver business results, not per-call computing prices.
Third, mature technical enablers. SenseTime’s upcoming NEO architecture multimodal model achieves a task completion rate of 94.3%, consuming only half the tokens of comparable mixed industry models. This robust technical foundation allows service providers to deliver consistent task results while containing internal costs, making task billing commercially viable.
5. Coexistence of Two Models: Complementary Rather Than Mutually Exclusive
Xu Li emphasized that task-based billing will not fully replace token billing; the two systems will coexist long-term to suit distinct scenarios.
Token billing retains value for low-level development, lightweight testing and exploratory prototyping, where flexible API adjustments are prioritized over predictable fixed costs. By contrast, standardized commercial products targeting consumer users and small-to-medium enterprises will widely adopt flat-rate task billing to meet demand for predictable expenses and streamlined usage. This layered dual-pricing system accommodates diverse use cases and fosters diversified industry development.
6. Industry Impacts: Reshaping Competitive Landscape and Accelerating AI Inclusiveness
The shift from token to task billing will generate far-reaching effects and advance high-quality industrial development:
First, a restructuring of AI service providers’ business models. Vendor competition will shift focus from raw computing prices to
task delivery efficiency, one-pass success rates and mature vertical industry solutions. This pressure will lift overall technical standards and service quality, pushing AI vendors to evolve from pure computing power suppliers to end-to-end solution providers.
Second, accelerated inclusive AI adoption. Ordinary users no longer need to master prompt optimization or context trimming to manage costs. AI becomes as intuitive to use as SaaS software, significantly lowering technical barriers for individual users and small merchants.
Third, the emergence of standardized vertical AI services spanning finance, legal services, content creation, office work, manufacturing and other sectors. Uniform pricing frameworks will form within niche verticals, deepening integration between AI and the real economy and supporting digital transformation across all industries.
7. Concurrent Viewpoints: AI as an Augmentation Tool, Balancing Safety and Governance
During the keynote, Xu Li shared additional perspectives defining the direction and bottom lines of AI development:
He framed AI as an
augmentation tool rather than a job replacement. While purely procedural systematic roles will be streamlined, AI will create a large number of new careers catering to "super individuals", energizing employment markets through harmonious human-AI collaboration.
He also proposed a three-tier AI security framework covering human-centric product design, data security during model training, and sandbox isolation for runtime environments. This multi-layered system safeguards secure AI deployment and upholds the industry’s safety baseline.
Furthermore, SenseTime has partnered with the United Nations to build an AI Governance Lab, advancing the formulation of global standards for responsible AI deployment, supporting healthy, orderly industry growth and refining worldwide AI governance frameworks.
8. Conclusion: Billing Reform Usher in a New Era of Inclusive AI
Xu Li’s proposed overhaul of AI billing models charts a clear new direction for the industry and illustrates the path for AI to move from niche technical circles to mainstream mass adoption. The shift from tokens to tasks essentially marks an industry-wide upgrade from technology-centric to value-centric operations.
With continuous technical progress and innovative business models, artificial intelligence will mature into universal infrastructure empowering all industries and all people. It will inject powerful momentum into social progress and unlock a new era of inclusive AI for everyone.