Dynamic Beating AI NewsFlash: Specializing in ultra-high-speed large models, Celeris has released a hybrid diffusion Agent model called Celeris-1 Magnus, based on Qwen3.8-27B. The official claim is that it can achieve over 500 Tokens/s in many scenarios.
On 97 banking customer service tasks for τ³-Banking, Magnus achieved a completion rate of 41.2% after enabling low reasoning, with a median time of 55 seconds. In the same scenario, GPT-5.6 Sol scored 38.1% in completion rate and took 79 seconds.
While traditional models generally generate one Token at a time, Magnus can generate multiple Tokens at once and then revise repeatedly. For content that needs to be generated strictly in order, it is processed serially. Therefore, each computation can advance more Tokens at once.
