Arcee AI specializes in small language models (SLMs) and provides tools for customers to train models as well as pre-trained models for domain-specific question-answering use cases. Its core technologies include Model Merging, which fuses multiple models into a single efficient model, and Spectrum, a targeted training technique that reduces training time and catastrophic forgetting. The company offers two delivery methods: Arcee Cloud (a hosted SaaS) and Arcee Enterprise (in‑VPC deployments favored by larger customers). Arcee positions SLMs as cost- and energy-efficient alternatives to very large models and says it can train GPT-3–like models for as little as $20,000. Pricing is sold as annual software-license contracts with inference charges, rather than purely usage-based pricing. Financially, Arcee reports $2 million in revenue, has 25 employees, plans to grow to about 50 within 18 months, and says there is a good chance it could turn profitable by early 2025. Arcee provides an end-to-end platform for building, fine-tuning, deploying and monitoring generative AI models inside a virtual private cloud to address enterprise security and ownership concerns. The product was developed by former Hugging Face and Roboflow engineers to let organizations train specialized language models on their own data within a secure compute environment. The company emphasizes adaptive training, superior fine-tuning and monitoring to mitigate privacy risks for highly regulated industries such as legal, healthcare, insurance and financial services. Arcee launched last February and is maintained by a Miami-based startup led by founders Mark McQuade, Brian Benedict and Jacob Salowetz. The startup has positioned itself against other tooling providers and large incumbents while highlighting demand from enterprises worried about transparency and security. Management says the capital raised so far will be used to expand the workforce, build out the platform and grow into new markets.

Total raised

$30M

Funding rounds

2

Latest round

Series A

Latest activity

Jul 2024