Fastino Trains AI models for cheap graphics games and just collected USD 17.5 million led by Khosl

Technological giants prefer to boast of trillion parameter AI models that require massive and expensive GPU clusters. But Fastino adopts a different approach.

The startup based on Palo Alto claims that he has invented a latest kind of architecture of the AI ​​model, which is deliberately small and specific to the task. Fastino says that the models are so small that they are trained in low GPU class with a value lower than $ 100,000.

The method attracts attention. Fastino secured $ 17.5 million of seed funds run by Khosl Ventures, the famous Investor Venture, Fastino, Fastino, only tells techniques.

This causes complete financing of the startup to almost $ 25 million. He collected $ 7 million in November last 12 months in the Przedseed round led by VC ARM M12 Microsoft and Insight Partners.

“Our models are faster, more accurate and cost a fraction for training, while exceeding the flagship models in specific tasks,” says Ash Lewis, general director of Fastino and co -founder.

Fastino has built a package of small models that he sells to corporate clients. Each model focuses on a specific task that a company might have, akin to the editors of confidential data or a summary of corporate documents.

Fastino does not reveal early indicators or users yet, but claims that its performance impresses early users. For example, because they are so small, his models can provide the entire answer in one token, Lewis told Techcrunch, showing technology, giving a detailed answer immediately in milliseconds.

It’s still a bit too early to seek out out if the fastino approach will gather. The AI ​​Enterprise space is crowded, and corporations akin to Cohere and Databicks also advertise artificial intelligence that leads in some tasks. And manufacturers of SATA models focused on the enterprise, including Antropic and Mistral, also offer small models. It is also no secret that the way forward for generative artificial intelligence for an enterprise is probably in smaller, more targeted language models.

Time can say, but Khosli’s early vote of confidence actually does not hurt. For now, Fastino says that he is focusing on building the latest AI team. He is directed to scientists from the best AI laboratories who are not obsessed with the construction of the largest model or beating comparative tests.

“Our employment strategy is very focused on researchers who may have a contradictory thought process in the construction of language models,” says Lewis.