Celeris-1 Magnus: Fast hybrid diffusion model for agentic work (celeris.ai)

Celeris has announced the launch of Celeris-1 Magnus, a cutting-edge hybrid diffusion model designed for rapid task completion in various applications, including chat and search interfaces. In comparisons against established models like GPT-5.6-sol and Gemini-3.7-flash on the τ³-bench banking, Magnus achieved a remarkable 41.2% solve rate and a median task completion time of just 55 seconds, setting a new standard for speed without sacrificing thoroughness. When the model is set to "thinking mode," it scores an additional 13.4 points, showcasing its capability for deeper reasoning when needed. This advance is particularly significant for the AI/ML community as it offers an OpenAI-compatible API, allowing developers to seamlessly integrate Magnus into their existing projects with minimal adjustments—essentially requiring only a change in the model identifier. With no waiting list and immediate access to API keys, Celeris-1 Magnus aims to enhance the user experience in real-time applications, making it a noteworthy competitor in the evolving landscape of AI task execution and agentic work.