Arcware today announced its emergence from stealth and introduced a platform for deploying and serving frontier artificial intelligence models.
Arcware was founded nine months ago with the goal of becoming the world’s first full-stack AI company. That begins with making existing open-weight AI models cheaper and more efficient to run. Arcware’s Atlas middleware architecture makes it possible to operate large language models on a significantly smaller hardware footprint while preserving their native weights. This makes serving these models more economically and environmentally viable as they continue to grow in size.
We are starting with DeepSeek V4 Pro, a 1.6-trillion-parameter mixture-of-experts model with 49 billion active parameters. Pricing is slated to begin at $0.22 per 1 million input tokens and $0.44 per 1 million output tokens: 50% below the current market rate. More details on performance are available here.
We plan to offer a direct Arcware API in the near future. Until then, we are working to make Arcware available through the AI gateways developers already use. DeepSeek is only the beginning. We plan to launch Kimi K3, GLM 5.2, and MiniMax M3 as soon as possible.
We are not content with being only an inference provider. Arcware Labs is also developing an open-weight foundation model and a coding assistant designed to be among the most affordable and efficient products on the market.
Throughout this work, Arcware will remain committed to building products that are more environmentally sustainable, economically viable, and private. From our headquarters in the San Francisco Bay Area, we operate locally hosted infrastructure designed to provide greater efficiency, reliability, and control.
Our mission is straightforward: to make artificial intelligence more affordable, more private, and better for the environment. We are excited to emerge from stealth, begin pursuing that mission publicly, and see what the next nine months bring.