Mercury 3: Diffusion at scale (www.inceptionlabs.ai)

Today marks the launch of Mercury 3, a groundbreaking diffusion-based large language model (LLM) that significantly outperforms existing models like GPT-6 Luna and Claude Haiku 5.5. With the ability to process over 1,000 tokens per second—4 to 8 times faster than its rivals—Mercury 3 stands out not only for its speed but also for its intelligence, boasting a parameter count more than 40 times larger than its predecessor, Mercury 1. Its scalable architecture addresses a critical question in the AI community about whether diffusion models could compete with autoregressive ones in complex applications, proving they can. Early-access partners are already integrating Mercury 3 into various real-time applications such as customer support, interactive education, and browser agents. These integrations highlight the model's capacity for low-latency, high-quality responses, essential for tasks requiring immediate feedback. For instance, in educational settings, teachers can provide instant responses to students, enhancing learning without sacrificing comprehension. With Mercury 3, businesses can expect not only improved performance but also cost efficiency, given its lower pricing structure. The model is now available through the Inception API Platform and is compatible with existing OpenAI API integrations, paving the way for wider adoption across industries.