Fastino Models AI is trained in graphics processing units for cheap games and just raised $ 17.5 million led by Khosla

Photo of author

By [email protected]


Technology giants love to boast models of artificial intelligence, a trillion parameter that requires huge and expensive GPU groups. but Fastino It takes a different approach.

The start -up company, which is based in Palu Alto, says it has invented a new type of artificial intelligence model, which is intentionally small and intended for the task. Fastino says the models are very small, as they were trained with low -game graphics processing units of less than $ 100,000.

The method is to attract attention. Fastino has obtained a $ 17.5 million seed financing led by Khosla Ventures, the first famous investor in Openai, exclusively informed Fastino Techcrunch.

This reaches the total financing to start the operation to approximately $ 25 million. She raised $ 7 million last November in the pre -seed tour led by ARM M12 M12 from Microsoft.

“Our models are faster, more accurate and cost a small training part while outperforming the main models in specific tasks,” says Ash Lewis, CEO of Fastino and co -founder.

Fastino has built a group of small models that it sells to institutional customers. Each model focuses on a specific task that the company may need, such as revising sensitive data or summarizing the company’s documents.

Fastino does not reveal early scales or users yet, but he says that its performance is tampering with the first users. For example, as it is very small, its models can provide a full response in one symbol, as Lewis told Techcrunch, where he showed the technique that gives a detailed answer once a millimeter again.

TECHRUNCH event

Berkeley, California
|
June 5


Book now

It is still a little early to know if the Fastino approach will continue. AI Enterprise space is crowded, as companies such as COHERE and Databrics are also promoting Amnesty International, which excels in some tasks. SATA makers also focus on the institution, including Antarbur and error, small models. It is also not a secret that the future of the Foundation’s artificial intelligence is It is possible in the most and most concentrated language models.

Time may say, but the early vote of confidence from a loser is definitely not hurt. Currently, Fastino says it focuses on building a team of advanced artificial intelligence. It targets researchers at Top Ai Labs who do not attack building the largest model or hitting standards.

“Our recruitment strategy focuses on researchers who may have a contradictory thinking process for how to build language models at the present time,” says Lewis.



https://techcrunch.com/wp-content/uploads/2025/05/v2-landscape.png?resize=1200,675

Source link

Leave a Comment