Liner has launched Liner Model API for companies running AI applications, saying it is designed to cut large language model costs by more than 50%.
The San Francisco-based group said the API routes each request to what it determines is the lowest-cost model able to complete the task, removing the need for developers to choose and switch models manually.
Businesses using large language models often send a wide range of work to expensive frontier systems, even when cheaper models could handle simpler requests. Liner said its system addresses that by assessing the expected quality and cost of candidate models, then selecting one for each request.
The company said the approach was developed through real-world use and further tested in controlled benchmarks covering model performance and cost efficiency. After deploying the routing system internally, Liner said its token spending in August fell by more than 50% compared with the first half of 2026.
The API is intended for tasks ranging from routine queries to coding, reasoning and research, the company said. Developers can use one interface across different workloads instead of maintaining separate model choices as tasks and prices change, it added.
Pricing model
Liner has set pricing for the API at USD $1 per 1 million input tokens, USD $6 per 1 million output tokens and USD $0.10 per 1 million cached input tokens. Based on publicly available pricing data, the service can cost at least 50% less than comparable models in the same performance tier, according to the company, although savings will vary depending on traffic patterns and workload mix.
The system uses a single-model selection method for each request rather than querying several models at once and comparing the results. Liner said this avoids the extra token spending associated with multi-model routing methods.
Benchmark comparisons published by the company measure the API against frontier models including Claude Sonnet 5 and GPT-5.6-Terra. Developers can review those results and estimate savings against their own spending through tools on Liner's developer platform, the company said.
CEO on launch
Luke Kim, Chief Executive Officer, Liner, framed the launch as a response to rising operational complexity in the AI software market. "Routing every query to a frontier model is the equivalent of spinning up a supercomputer just to solve basic arithmetic," Kim said. "Liner Model API eliminates the constant operational headache developers face when trying to benchmark, price and swap out models as they hit the market."
Liner describes itself as an evidence-first AI research platform focused on outputs that users can verify through citations and source traceability. Its tools are used by students, educators, professionals and other knowledge workers who need auditable responses for research and decision-making, according to the company.