The cost of model calls is rising rapidly after the large-scale enterprise deployment of the generated AI. Emgram, which was established for only eight months, announced the completion of $98 million in financing to try to reduce token consumption and improve response efficiency through an enterprise-oriented “memory layer” product.
Financing and client lists
The current round of financing was attended by General Catallyst, Kleiner Perkins and Sequoia, and the co-founder of OpenAI, Andrej Karpathy, was among the investors. According to Engram, funds will be used mainly for computing inputs and talent recruitment.
The company currently has only 13 employees, but it has access to clients such as Microsoft, Notion and the Law AI Initial Company Harvey. For a start-up enterprise with a short start-up time, such a list of clients indicates that its products have entered the enterprise ' s trial and procurement horizon.
Main reduction token expenditure
Engram has positioned itself as "learning memory" for AI. The core idea is to allow the model to preserve work processes, context and historical information within the organization, thereby making more rapid use of relevant elements in follow-up questions and answers and reducing the costs of repetitive reasoning and lengthy context.
According to the company, in some of the missions, its models could be as much as 100 times as token, with comparable or better results than the front laboratory model. Token is the base unit of the AI query running, and the higher the call, the higher the enterprise usually spends.
Cut into model memory panel
Dan Biderman, co-founder and Chief Executive Officer, argues that the problem with many of the current models is not the lack of understanding, but the lack of stable memory. The longer the context, the easier the model is to be slowed down by additional information, and the higher the cost to enterprises of more retrieval, reading and reasoning.
He stated that Engram did not claim to be better than the OpenAI or Anthropic model in all its capabilities, but rather placed more emphasis on efficiency advantages in a professional setting. This financing also shows that AI entrepreneurship companies are moving from simply pursuing stronger models to cost control and business landing capacity.