Palo Alto – based artificial intelligence startup Fastino just raised $17.5 million in seed funding, led by Khosla Ventures. The company has raised close to $25 million so far, including an earlier $7 million pre-seed round raised in November 2024 led by Microsoft’s venture arm M12 and Insight Partners. The new funding will be used to grow the company’s research team and continue to develop the company’s suite of task specific AI models into enterprise ready products.
Instead of going after the same commercial play as most startups building huge, general purpose language models of the type GPT or Gemini, Fastino is doing something very different. The company begins by creating lightweight, efficient task specific language models (TLMs) that perform one thing and one thing only well, such as summarization of documents or redact of sensitive data and well. Not only are these models fast, they are also trained on inexpensive gaming GPUs, orders of magnitude cheaper than their larger brethren.
Ash Lewis, Fastino’s CEO, says the company is focused on the need to create ‘fast, cheap, and accurate’ AI. Their TLMs, he said, routinely produce results in just a single token, making them vastly responsive for business needs. This is a big shift from the direction of building monolithic models that do not require much computational power and infrastructure. Actually, Fastino are training their models on an off-the-shelf cluster of gaming GPUs that total under $100,000 in value, proving their love for the innovation at an affordable price.
Compact and efficient models are focused on in order to help shorten development cycles and make easier to integrate in enterprise systems. Pilot projects with sensitive data processing have successfully been applied to Fastino’s current models. It’s not clear what specific clients, or hard numbers, are using the models, but those familiar with the technology say they are exceeding project estimates on metrics like latency and precision.
At this time, this strategic direction is at the epicenter of interest from investors and researchers who are increasingly wary of environment and financial cost of training large scale models. According to Fastino, his main focus on real world applications as well as the minimal hardware requirements of Myria is a democratizing force in AI development. This positions the company to provide scalable, tailored AI tools to a host of businesses without incurring the high costs and complexity of large model training.
Fastino will continue scaling its AI capabilities and recruit top talent, it plans to expand into many more enterprise use cases making Fastino a unique player in the evolving world of AI.
