Liner Launches

Liner announced on the 15th the launch of its AI orchestration solution, "Liner Model API." The company cited the cost burden involved in building enterprise AI systems as the reason for the launch.

Liner explained that many companies are currently processing all user queries in bulk with a single expensive frontier model to ensure answer quality. But the company said the actual incoming queries vary widely in difficulty.

The company also said that handling every query with a top-tier model creates a structural limitation that increases infrastructure costs. It therefore introduced "Liner Model API," which automatically assigns the appropriate LLM based on query difficulty.

The solution assigns a high-performance model to difficult queries and a lightweight model to general queries. Liner said the goal is to maintain answer quality while improving cost efficiency, and it said the approach can cut token costs by more than 50%.

Liner said it designed Liner Model API for enterprise demand from companies weighing a choice between quality and cost. It said those companies are concerned about the quality issues of low-cost lightweight models and the adoption costs of frontier models. The company said Liner Model API maintains answer accuracy at the level of top-tier models while cutting token costs by more than 50%. It added that no separate logic implementation is needed and that the service can be applied immediately by simply replacing the existing API.

The company said the core of the API is its precision orchestration technology. Liner said it operates by dividing models according to query difficulty, assigning difficult queries involving coding, math, and complex reasoning to top-tier models to ensure accuracy. According to an evaluation of Liner's own real-world usage data, about 43% of queries are assigned to high-performance models. The remaining 57% of general queries are routed to efficient models, which the company said helps improve cost savings while maintaining accuracy.

Liner said its key competitive advantages in making this operation possible are its proprietary model evaluation system and its experience verifying performance in real-world environments. It also said it has built various LLM evaluation systems while operating AI agent products it runs, including Liner Search, Liner Scholar, Liner Light, and Liner Finance. The company said this evaluation system and real-world verification experience form the basis for its model allocation operations.

Liner said that during the performance verification process, it built a de-identified representative query dataset to reflect actual user queries. It then analyzed usage patterns such as everyday search, work tasks, and academic research, and said it optimized the model assignment algorithm based on that analysis. Liner said the algorithm optimization helped secure service stability and performance.

Liner CEO Kim Jin-woo said that processing every question only with a top-tier model is similar to mobilizing a supercomputer for simple calculations every time. He also said Liner Model API will solve the inconvenience of developers having to repeatedly verify and replace the performance and pricing of new models as they appear. He added that the company aims to create an environment where enterprises can move beyond cost pressure and model-selection fatigue and focus on innovating core business operations.

Source: IT DAILY · Yang Seung-gap
Original: https://www.itdaily.kr/news/articleView.html?idxno=241622

References

• https://www.itdaily.kr/news/articleView.html?idxno=241622