Celeris Unveils Celeris-1 Magnus Hybrid Model

Celeris posted an announcement introducing Celeris-1 Magnus as a hybrid diffusion model derived from qwen3.8-27b. The company stated the model targets agentic workloads. Its post reported that Magnus reached 41.2 percent at a 55-second median on the τ³-bench banking task. The same message listed GPT-5.6-sol at 38.1 percent with a 79-second median. Celeris positioned these outcomes as evidence of faster performance suited to agentic work. The announcement included a link to its site for further details. No other posts in the packet provide independent confirmation of the results.