Celeris, which focuses on ultra-high-speed large models, has launched Celeris-1 Magnus, a hybrid diffusion agent built on Qwen3.8-27B. The company says the model can run at more than 500 tokens per second in many scenarios. On the τ³-Banking benchmark, which includes 97 bank customer-service tasks, Magnus achieved a 41.2% completion rate with low reasoning enabled and a median completion time of 55 seconds. In the same test, GPT-5.6 Sol recorded a 38.1% completion rate and a median time of 79 seconds. Celeris says the speed comes from a different decoding approach. Instead of writing one token at a time, Magnus produces multiple tokens in a single pass, then revises them iteratively. Content that needs strict ordering is still handled serially, which lets one computation step advance more tokens than a conventional token-by-token approach. The test figures come from the company's own testing, as reported by BlockBeats.
Celeris, a developer focused on ultra-high-speed large models, has released Celeris-1 Magnus, a hybrid diffusion agent built on Qwen3.8-27B. The company says the model can exceed 500 tokens per second in many scenarios.
Banking benchmark: 41.2% completion rate
On τ³-Banking, a benchmark made up of 97 bank customer-service tasks, Magnus posted a 41.2% completion rate with low reasoning enabled and a median time of 55 seconds. GPT-5.6 Sol finished the same evaluation at 38.1% with a median of 79 seconds.
Diffusion-style decoding
Traditional models typically generate one token at a time. Magnus instead produces multiple tokens in one pass and then revises them iteratively. When a sequence must follow a strict order, the model handles those parts serially. That design allows a single computation to push more tokens forward than conventional token-by-token generation.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
