Reasoning about agents’ beliefs with varying degrees of confidence is crucial in many AI domains, especially when dealing with uncertain, incomplete, or evolving information. Graded Dynamic Epistemic Logic (GDEL) extends standard Dynamic Epistemic Logic (DEL) by introducing a quantitative notion of belief: an agent believes a proposition \(\varphi \) with strength n if \(\varphi \) holds in accessible worlds whose cumulative plausibility meets or exceeds n. This framework enables the modeling of multi-agent systems with graded and higher-order knowledge. While expressive, GDEL suffers from scalability issues due to the combinatorial explosion of states in DEL-style Kripke structures. An effective and compact symbolic representation is thus required to handle graded knowledge updates efficiently. In this paper, we propose a symbolic theoretical framework for GDEL based on pseudo-Boolean formulas (PBFs), extending prior work on symbolic representations for DEL and Probabilistic DEL (PDEL). In our approach, possible worlds are encoded as Boolean valuations, while graded beliefs are represented as threshold constraints over weighted transitions. Dynamic updates, such as public announcements, are modeled via symbolic restriction and transformation of these structures. To enable practical reasoning, we instantiate this framework using Algebraic Decision Diagrams (ADDs)–a well-known data structure from knowledge compilation—to efficiently encode and manipulate weighted relations. Our framework remains modular and can accommodate alternative representations, such as Semiring-Labelled Decision Diagrams (SLDDs) or Affine Algebraic Decision Diagrams (AADDs). We validate our approach through Python-based experiments on the cooperative card game Hanabi, a benchmark for reasoning under uncertainty and hidden information. Our symbolic GDEL engine supports scalable multi-agent reasoning over thousands of possible worlds, while preserving logical clarity and computational efficiency. This work lays the foundation for integrating graded belief models into symbolic epistemic planning and reasoning in uncertain multi-agent environments.