a lightweight hybrid reasoning MoE model with 7.9B total parameters and only 1.3B activated parameters per token. It is designed to deliver strong reasoning and agentic capabilities under a small inference compute footprint, making advanced model capabilities more accessible for local and resource-constrained deployment.
I haven’t tried it yet, but I tested one of their previous ones with CPU interference on an Intel n97, and it was one of the best in terms of t/s performance and also prompt response quality on the metrics I tested against. When this run on llama.cpp I will try and give it a test.