Third-Party LLMs and Usage Multipliers Reshape an AI Office Assistant

    An AI office assistant has expanded beyond its proprietary models by adding three third-party language models and a usage multiplier system tied to actual consumption. This article explains how model selection, transparent credit costs, time-based discounts, and model-specific multiplier reductions affect everyday office workflows and enterprise AI budgeting.

    From a Single Model to Multi-Model Collaboration

    As large language models become more widely used in workplace applications, organizations are placing greater emphasis on balancing three factors: model quality, response speed, and usage cost. Office AI products traditionally relied on proprietary models, which resulted in relatively fixed capabilities and cost structures. However, as the model services market continues to evolve, it has become increasingly difficult for any single model to deliver the best results across every type of task.

    A recent product update from an AI office assistant platform expanded its supported model range from proprietary models to third-party models. At the same time, the platform introduced a usage adjustment mechanism based on actual consumption. This represents a significant shift from a closed model stack toward a more open model ecosystem for workplace AI.

    Core Update: Third-Party Models and Model Selection

    Previously, the office assistant offered only two proprietary models: one designed for lightweight, high-frequency tasks and another focused on more complex reasoning. Following the update, three third-party models were added to the available model list, covering different parameter scales and reasoning capabilities.

    Users can switch between proprietary and third-party models according to the requirements of a task:

    • Lightweight tasks—such as summarization, information extraction, and format conversion—can use low-multiplier models to reduce credit consumption.
    • Medium-complexity tasks—such as document understanding, code generation, and multi-turn conversations—can use balanced models that offer a compromise between quality and speed.
    • Complex tasks—such as long-form analysis, logical reasoning, and structured output generation—can use more capable models to improve response quality.

    The central value of this design is that it puts model selection directly in the user’s hands. Under a traditional single-model approach, users must accept the capabilities of the model built into the product and cannot adjust resource allocation according to the task. With multiple models available, users can view each model’s intended use and consumption coefficient directly in the product interface. Model selection becomes a needs-based decision rather than a matter of guesswork.

    The platform had previously made third-party model configuration available to customers on team and enterprise subscription plans. In that phase, organizations did not need to purchase services from model providers separately or apply for independent APIs. Usage was deducted directly from their existing AI credits, creating a unified account and billing structure.

    The latest update extends this capability to a broader user base and introduces a limited-time promotional program for all eligible users.

    Understanding the Usage Multiplier Mechanism

    Once multiple models are available, differences in operating cost become unavoidable. Models consume different amounts of resources when completing the same task. These differences are related not only to parameter size but also to inference architecture and computing efficiency.

    The platform addresses this issue by displaying a usage multiplier for each model on the model selection page.

    A Pricing Coefficient, Not a Quality Rating

    A usage multiplier can be understood as a pricing coefficient applied when different models complete the same task. It does not change the complexity of the task itself. Instead, it determines how quickly the user’s available credits are consumed.

    For example:

    • A lightweight model with a multiplier of 0.3 consumes 30% of the baseline credits for the same task.
    • A high-performance model with a multiplier of 1.5 consumes 1.5 times the baseline credits.

    The main benefit of this mechanism is that it makes cost visibility part of the user experience. Users do not need to estimate which model is more economical based on experience or trial and error. Before submitting a task, they can understand the expected consumption and make a more informed trade-off between quality, speed, and cost.

    Dynamic Multipliers as an Operational Tool

    The multiplier system also gives the platform a flexible way to adjust consumption for specific models or time periods without changing its overall pricing structure.

    This is similar to the way cloud computing services charge for different instance specifications. Computing resources vary in cost depending on their configuration, while multiplier coefficients provide a mechanism for managing supply, demand, and resource allocation.

    Two-Part Limited-Time Promotion

    The update also introduces a limited-time promotion with two components.

    Time-Based Discounts

    During designated periods on weekdays and holidays, participating models are available at discounted credit consumption rates.

    This structure reflects common usage patterns in office environments. During peak working hours, demand is concentrated, while discounts can encourage users to shift some workloads to other periods. During off-peak hours, computing resources may be less heavily used, allowing the platform to improve utilization by offering lower consumption rates.

    Reduced Multipliers for Selected Models

    The platform has also reduced the multipliers for two specific models: a proprietary lightweight model and a third-party model designed for efficient reasoning.

    Both models are positioned for frequent, relatively lightweight office tasks, including drafting emails, organizing meeting notes, and summarizing documents. The effect of a lower multiplier becomes more significant as usage frequency increases. For teams that perform dozens or even hundreds of AI tasks each day, the long-term difference in credit consumption can be substantial.

    Practical Impact and Model Selection Guidance

    Scenario 1: High-Frequency, Lightweight Tasks

    For routine tasks such as message summarization, information classification, and content rewriting, users should generally prioritize low-multiplier models.

    These tasks typically require limited reasoning depth. In many cases, a lightweight model can provide satisfactory output quality while significantly reducing credit consumption.

    Scenario 2: Medium-Frequency, Moderately Complex Tasks

    For tasks such as technical document drafting, data analysis assistance, and multi-turn business conversations, users should select models based on their immediate requirements.

    During promotional periods, participating models may provide a more favorable cost structure. Outside those periods, users can temporarily switch to higher-multiplier models when a task requires stronger output quality or more advanced reasoning.

    Scenario 3: Infrequent, High-Complexity Tasks

    For research reports, complex business proposals, and in-depth analysis of long documents, users should retain access to higher-multiplier models.

    Although these tasks consume more credits per request, they generally represent a smaller share of total usage. Their effect on overall costs can therefore remain manageable when high-capability models are reserved for situations where they provide clear value.

    Industry Perspective: From Model Binding to Model Routing

    The broader significance of this update is that it reflects the transition of office AI products from single-model binding to multi-model routing.

    In the past, workplace tools often depended on a single model provider, which meant their upper capability limit was closely tied to that provider’s development cycle. With multiple models integrated into the same product, the platform can adapt to different task requirements without changing the user’s familiar workflow.

    For enterprise users, this shift offers several direct benefits:

    • Greater cost transparency: Multiplier labels help IT administrators estimate AI usage costs more precisely.
    • More flexible model selection: Different teams can choose model combinations that match their business requirements instead of accepting a single, standardized configuration.
    • A more open ecosystem: Third-party model support allows the platform to extend beyond its proprietary capabilities and follow advances across the broader model market.

    Conclusion

    The combination of third-party model access and usage multipliers is essentially an effort to make costs measurable and user experience more adjustable.

    For individual users, it lowers the barrier to trying different models. For administrators, it provides a clearer basis for tracking and planning AI consumption. For the industry, it demonstrates how workplace AI products are evolving from simply delivering features to orchestrating computing resources.

    Competition among language models will continue, but the most important value for end users is not necessarily the standalone capability of one particular model. It is whether a product provides enough flexibility to select the right model and enough transparency to understand the resulting cost.

    From this perspective, the move toward multi-model access, visible multipliers, and time-based incentives is an important development to watch.