ZGCM-1: Efficient 7B Open Foundation Model for Math and Search
September 10, 2026
ZGCM-1 is a 7B dense model using interleaved gated sliding-window and full attention with an FP8 Muon optimizer. It utilizes a 256K context window and reformulates interaction traces into Markov Decision Processes to overcome parametric limits through active tool use.
HOW THIS AFFECTS YOU
●
builderYou can deploy a highly efficient 7B model capable of long-context tool use.
●
researcherThe combination of MDP mid-training and FP8 Muon offers a new recipe for compact model efficiency.