DeepSeek-R1T-Chimera: Happy 1B tokens/day!

May 29th, 2025

Our R1T Chimera 671B research prototype, constructed from DeepSeek R1 and V3-0324, has now been in the wild for one month.

Usage reached up to 1.35 billion processed tokens per day on OpenRouter, mostly for chat and the coder apps Roo Code and Cline.
On the underlying Chutes serverless AI platform, the Chimera peaked as being the third most-popular model. It was used in over 300k runs per day, processing 4.87 billion tokens/day. In total since coming online, it processed about 100 billion tokens.
Typically, it runs on about 25 Chutes instances of 8xH200 each, i.e. 200 x H200 GPUs.

Thanks to the community for downloading it 4,500 times from HuggingFace, to DeepSeek for creating the parent models, and to OpenRouter and Chutes for hosting it.

PS1: TNG with about 1,000 people processes 100M tokens/day. Thus, 5B tokens/day correspond to about 50,000 TNG-type people.

PS2: A cousin is in the works.