Liner Launches 'Liner Model API' to Cut LLM Costs by More Than 50%
TECHWORLD ·
✦ AI Summary
Liner announced on the 15th that it had launched the 'Liner Model API.'
The API is an AI orchestration solution that automatically assigns the most suitable LLM based on question difficulty and type.
Liner said it can cut LLM token costs by more than 50% while maintaining answer quality.
Liner announced on the 15th that it had launched the 'Liner Model API.' The 'Liner Model API' is an AI orchestration solution equipped with a function that automatically assigns the most suitable LLM based on question difficulty. Rather than processing all queries with a single high-performance model, the API aims to reduce token costs while maintaining answer quality.
The API applies a structure that assigns models separately according to query difficulty and type. It assigns a top-tier model to high-difficulty queries such as coding, math, and complex reasoning, and a relatively lower-cost model to general knowledge queries. Liner said this can maintain answer quality at the level of top-tier models while cutting LLM token costs by more than 50%.
The implementation method involves replacing only the API in an existing system. As a result, it can reduce the burden of having to implement separate, complex logic. The key element is orchestration technology that determines the most suitable model for each question type.
Liner said it had built a system to compare and evaluate the performance of multiple LLMs in the course of operating its AI agent services. The company said it implemented orchestration technology that determines the most suitable model for each query based on that system. The services Liner operates are 'Liner Search,' 'Liner Scholar,' 'Liner Lite,' and 'Liner Finance.'
Liner used query data accumulated in real service environments in addition to public benchmarks. Based on a de-identified representative query dataset, Liner verified model performance by question type and refined its model assignment algorithm.
According to Liner's internal evaluation, about 43% of queries were assigned to high-performance models. Queries requiring coding, math, and complex reasoning are handled by top-tier models, while the remaining 57% of general queries are assigned to relatively efficient models.
Liner's main target is companies that are weighing the high cost of frontier-grade models against the quality of lightweight models. Liner said it focused on reducing the burden on companies of directly comparing and replacing the performance and pricing of each model, and on providing a cost-efficient LLM operating environment.
With this product launch, Liner plans to expand its business scope from its existing AI agent services into the B2B AI infrastructure space. Liner will provide, in API form, the model evaluation experience it has accumulated while operating its own services so that outside companies can use it.
Kim Jin-woo, CEO of Liner, said it is inefficient to rely only on high-performance models for every request, comparing it to using a supercomputer for simple calculations. He added that the Liner Model API can ease the burden of verifying the performance and cost of new models, and also reduce the burden on developers who have to check a model's performance and cost and then replace it every time a new model is released.
Source: TECHWORLD · Kim Seung-ki
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406945
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.