Fastino Trains AI models of cheap games for games and just raised $ 17.5 million led by Khosla

Technical giants like to boast about the AI ​​trillion-parameter models that require massive and expensive GPU clusters. But Fastino takes a different approach.

Starting, based on Palo Alto, says it has invented a new type of AI model architecture that is deliberately small and specific to the task. The models are so small that they are trained with low-end playing graphic processors worth less than $ 100,000, says Fastino.

The method attracts attention. Fastino has provided $ 17.5 million in seed funding, led by Khosla Ventures, the famous OpenAi investor for the first business, says Fastino exclusively to TechCrunch.

This leads to the total funding of the startup to nearly $ 25 million. He raised $ 7 million last November in a round before the seeds led by Microsoft and Partners Insight’s VC Arm M12.

« Our models are faster, more accurate and cost a faction to train while superior to the leading models of specific tasks, » says Ash Lewis, CEO and co-founder of Fastino.

Fastino has built a set of small models it sells to corporate customers. Each model focuses on a specific task that the company may need, such as editing sensitive data or summarizing corporate documents.

Fastino does not yet reveal early indicators or users, but says its performance awakens early users. For example, since they are so small, his models can deliver a whole answer to a marker, Lewis told TechCrunch, showing the technology, giving a detailed answer at a time in milliseconds.

It is still a little earlier to say whether the Fastino approach will get involved. The Enterprise AI space is overcrowded, with companies like Cohere and Databricks also announced AI, which is distinguished by certain tasks. And manufacturers of SATA models focused on Enterprise, including anthropic and mitral, also offer small models. It is also no secret that the future of generative AIA for the enterprise is probably in less, more focused linguistic models.

Time may say, but the early vote of Hosla’s confidence certainly does not hurt. So far, Fastino says he is focused on building an avant -garde AI team. It is aimed at researchers in the best AI laboratories that are not obsessed with the construction of the largest model or beating the indicators.

« Our hiring strategy is very focused on researchers, who may have a counter -foreign mental process on how language models are being built at the moment, » Lewis says.