Microsoft Research released GraphRAG 1.0 with configuration and data-model changes

Microsoft Research dated GraphRAG 1.0 to December 16^{[source 1]}

On December 16, 2024, Microsoft Research announced the official release of GraphRAG 1.0. The post described ergonomic and structural improvements after a pre-release period, including easier setup for new projects. A 1.0 label is a versioning event, not a statement that a generated graph is complete, that entity extraction is correct, or that a system can answer questions safely over sensitive material. The dated record supports a release milestone and its stated engineering focus.

The project treated configuration as part of the interface contract^{[source 1][source 2]}

The introduction described refactors intended to make configuration and project setup easier to manage. The immutable v1.0.0 release adds concrete context: it lists a migration notebook, community parent IDs, a separate community workflow and factory-class cleanup. That is useful operational detail because a migration is not decoration in a retrieval pipeline; it changes the reproducibility work needed around models, storage, prompts and indexing.

Data-model changes made migration evidence separate from a package upgrade^{[source 1][source 2]}

The v1.0.0 release’s parent-ID and migration-notebook notes make migration evidence separate from a package upgrade. Such changes can require a fresh index or a migration path depending on what a user has stored. They do not say a graph is more truthful than its source documents. A retrieval system still needs source ownership, deletion handling, provenance and tests for whether a query returns the supporting record rather than a plausible but unsupported summary.

The December release did not settle the quality of downstream retrieval^{[source 1][source 2]}

GraphRAG 1.0 made a concrete software event out of an earlier research-oriented project: an official release with changed configuration and data boundaries. It did not make GraphRAG a substitute for evaluation. Results can vary with document extraction, graph construction, model choice, indexing parameters and query design. This article does not use later versions, current documentation or benchmarks to enlarge the December claim. It records the release and the practical migration boundary described by first-party materials. A team evaluating a graph result still needs to inspect linked source passages, test ambiguous entities and measure behavior when the corpus changes. A version number cannot supply that evidence. It also needs a documented decision for out-of-date source records: whether to reindex, exclude them, or label answers as incomplete. That question is operational rather than solved by graph topology.

Migration evidence should include the old and new configuration, the source snapshot and a sample of queries before any claim that results remained comparable. Reindexing may change entity extraction and relationships even when the original source documents have not changed.

The December sources also make clear that release provenance and content provenance are different records; both must be retained for a reproducible review.