Mercury 2.5 (www.inceptionlabs.ai)

Today marks the launch of Mercury 2.5, a significant upgrade over its predecessor, Mercury 2, designed for low-latency, low-cost AI applications. With a remarkable 40% increase in intelligence and a robust capacity to process 1,107 tokens per second, Mercury 2.5 stands as the largest diffusion language model (LLM) available, matching the capabilities of cutting-edge models like GPT-5.6 Luna and Gemini 3.5 Flash-Lite. Notably, it supports extensive context handling with up to 260,000 tokens while being priced attractively at $0.04 per million input and $0.15 per million output for a limited time. This upgrade is particularly significant for industries relying on real-time AI capabilities, including search, voice technologies, and coding applications. Real-world implementations have already demonstrated impressive advancements: companies using Mercury 2.5 reported latency improvements from several minutes to mere seconds in voice processing, and coding tasks saw an 82% reduction in latency and a 90% cost decrease. Additionally, innovations like Mercury Voice and Mercury Router enhance the user experience by optimizing response times for voice interactions and efficiently routing requests to the most suitable models. This launch highlights rapid advancements in AI model architectures and their increasing readiness for production across various domains.