Arcee AI, a new Miami-based startup, is already having an impact in the artificial intelligence space. They’ve created Trinity, a pioneering open-source language model that raises the bar. Such is the new state of the art foundation model with a staggering 400 billion parameters. It’s the largest open-source foundation model ever trained and released by a U.S. company. Arcee AI’s quick development within a 6-month timeframe shows the ambitious drive to compete with serious contenders such as Meta’s Llama.
Lucas Atkins, director of policy at Arcee AI underscored that the company’s approach was “very intentional.” Such focused and strategic planning helped position them to attract a globally competitive AI lab in record swift time. Before diving into large language models like LLMs, Atkins honed his skills by developing voice agents for the auto sector. It was this experience that served as a launching point for his deft touch with game-changing AI development.
With the release of Trinity, we feel this is an exciting and new chapter in Arcee AI. The model as per the Apache 3.0 license under which it is licensed will continue to stay open, available and amenable to public development forever. This decision aligns with the company’s goal of providing a permanent alternative to the proprietary models that dominate the market. Arcee would like to focus on U.S. companies that may be searching for open models. Unlike Chinese alternatives, these models don’t come with the commercial limitations which typically accompany them.
“Our aim is to woo U.S. companies of all sizes away from choosing open models from China,” stated Atkins. He highlighted the value of providing an open-weight model that’s best-in-class in performance. “Ultimately, the winners of this game, and the only way to really win over the usage, is to have the best open-weight model,” he added.
Whether it’s coding tasks or complex multi-step processes, Trinity’s capabilities are built for applications from top to bottom. The biggest version of Trinity will come in three different formats, providing users with flexibility to use Trinity in ways that best fit their unique needs. Early benchmarks show that Trinity is on par with Llama for use cases including coding, maths, and reasoning.
Mark McQuade, founder and CEO of Arcee AI, highlighted their innovative approach to the field. “I think there’s less than 20 companies in the world that have ever pre-trained and released their own model,” he remarked. Having been one of the original employees of the open-source company HuggingFace, McQuade has knowledge about what builds the best AI models.
Arcee AI about making a multi-million dollar bet on training Trinity, investing nearly $20 million. This massive project was powered by the impressive 2,048 Nvidia Blackwell B300 GPUs. This huge resource investment is just one sign of their drive to improve quality and performance in a very competitive market. So far, Arcee has been able to crowdfund nearly $50 million to fund its mission.
While leveraging the power of advanced language models, Arcee AI is developing new ways to compute and understand AI. The original business is in the process of producing a vision-based model and is looking to produce a speech-to-text version in early 2024. These improvements are designed to expand the use cases for their tech and strengthen their foothold in the AI market.
Though still in early stages, McQuade has definitively established Arcee AI’s mission in light of today’s market demand. “Arcee exists because the U.S. needs a permanently open, Apache-licensed, frontier-grade alternative that can actually compete at today’s frontier,” he stated. What makes Arcee unique is that it focuses on open-source principles. Atkins slammed models such as Llama for being not truly open source because they run on a Meta-controlled license.
“Llama can be looked at as not truly open source as it uses a Meta-controlled license with commercial and usage caveats.” – Lucas Atkins
