[DigitalToday reporter Jin-ho Lee] AI agent startup Liner unveiled an AI orchestration solution that automatically assigns suitable large language models (LLMs) based on user queries to reduce costs.

Liner said on Tuesday it launched the 'Liner Model API'. It is designed to cut LLM token costs by more than 50 percent while maintaining answer quality at the level of high-performance models.

Companies' AI systems currently process all queries with high-performance models even though questions vary in difficulty, creating unnecessary infrastructure costs, the company said. Liner Model API analyses query difficulty and type to automatically assign an appropriate LLM. It applies high-performance models to difficult queries such as coding, math and complex reasoning, while using more cost-efficient models for general queries.

Another feature is that it can be applied by replacing an existing API without separately implementing complex logic. Liner developed the orchestration technology based on its in-house model evaluation system and data accumulated during real service operations.

It verified performance by model using not only public benchmarks but also real users' query data. In Liner's in-house evaluation, about 43 percent of all queries were assigned to high-performance models. Queries requiring coding, math and complex reasoning were handled with top models, while the remaining 57 percent were processed with more cost-efficient models.

Liner said this can reduce token costs by more than 50 percent while maintaining answer accuracy.

Liner CEO Jin-woo Kim (김진우) said, "Handling every question only with top models is like bringing in a supercomputer for every simple calculation," adding, "Liner Model API will solve the hassle of developers having to verify and replace the performance and price of newly released models each time."