en.Wedoany.com Reported - AI infrastructure company Neurometric AI has launched an automated token engineering platform and announced the completion of a $4 million funding round earlier this year. The platform is designed to help enterprises manage the cost and performance of agent workloads.
As enterprises transition AI agents from experimentation to production, a single workflow can generate dozens of model calls. Many enterprises still send every task to frontier models, even when smaller, cheaper models can achieve the same or better results. Neurometric evaluates each call individually, modifies prompts as needed, and routes each task to the most cost-effective model that meets the required performance threshold; when no suitable option exists, it creates a purpose-built small language model. For simple, high-volume workloads, the platform automatically generates specialized small language models to optimize task speed and cost.
"Over the past year, enterprises have proven that AI agents can perform increasingly complex work. Now they need to prove that the economics still make sense when these agents run at scale," said Rob May, CEO of Neurometric. "Every model call is also a pricing decision, and these decisions compound across an agent's workflow. Token engineering provides enterprises with a way to control costs without sacrificing quality."
Currently, enterprises rely on manual testing and individual point solutions to decide which models should handle different AI tasks. As new models enter the market and pricing, speed, and performance change, these choices can quickly become outdated. Neurometric integrates model routing, small language model creation, and access to a marketplace of pre-trained task-specific small language models (SLMs) into a single platform. Its Task Endpoint Manager evaluates incoming requests based on continuously updated model performance and pricing data, then routes each task according to the customer's accuracy, cost, and latency requirements. When no existing model meets these requirements, its Auto-SLM Creator builds and provides a small language model for the specific task. The platform's SLM marketplace also allows customers to access models already developed for common and repetitive workloads.
In early customer engagements, models routed or created through Neurometric have achieved accuracy rates up to 20 percentage points higher than frontier models, while also reducing cost and latency.
Neurometric completed a $4 million funding round earlier this year, with participants including Betaworks, ex-Ante, Everywhere.vc, Encoded, Vermillion, Abstraction, and Mu Ventures, as well as angel investors including All-In Podcast co-host Jason Calacanis and Hubspot CTO Dharmesh Shah. The funds will be used to expand the engineering and AI research teams and provide more optimization tools for its core platform.
Neurometric positions token engineering as a discipline that determines how each task in an AI workload should be completed based on required quality, cost, and speed. Unlike prompt engineering, which focuses on improving instructions given to a model, token engineering decides which model should receive the task first and whether a more specialized model should be created to handle it. As enterprises deploy more AI agents, individual workflows generate more model calls, and the number of available models continues to grow, demand for this capability is expected to increase. Currently, the platform is live at neurometric.ai, and the company plans to meet with customers, investors, and media during the AI Engineer World's Fair in San Francisco from June 29 to July 2.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
