FalkorDB Ships GraphRAG SDK 1.0 - Ranks #1 on GraphRAG-Bench Across All Four Task Types

FOR IMMEDIATE RELEASE
April 29, 2026

Open-source framework outperforms Microsoft GraphRAG, LightRAG, and seven other systems. Runs on a MacBook Air. Costs $0.001 per query.

Tel Aviv, Israel - April 29, 2026 - FalkorDB today released GraphRAG SDK 1.0, an open-source, LLM-agnostic framework for building production-grade knowledge graph pipelines. In an independent evaluation on GraphRAG-Bench - the field's leading multi-domain benchmark - the SDK ranked first overall across all four question types: Fact Retrieval, Complex Reasoning, Contextual Summarization, and Creative Generation.

Results on the GraphRAG-Bench leaderboard (ACC ×100):

Rank System Novel (Multi-Doc) Medical (Single-Doc) Overall
#1 FalkorDB GraphRAG SDK 63.73 75.73 69.73
#2 G-Reasoner 58.94 73.30 66.12
#3 AutoPrunedRetriever 63.72 67.00 65.36
The closest competitor scores 14 points lower overall. All benchmark runs were executed on a MacBook Air (Apple M3, 24 GB) using GPT-4o-mini - no GPU cluster required.

The problem GraphRAG SDK 1.0 solves

Standard RAG pipelines degrade at scale. Vector-based retrieval returns semantically similar chunks, not connected knowledge. Multi-hop questions trigger dozens of LLM calls. Token costs run 3-5× higher in production than in test environments. The result: a system that appeared accurate at 10 documents and fails at 1,000.

Earlier GraphRAG implementations introduced new failure modes: brittle extraction pipelines that hallucinate entities, graph construction overhead that consumed compute before answering a single question, and defaults that required months of tuning.

GraphRAG SDK 1.0 was built to replace all of that.

What ships in 1.0

Modular API - pipelines teams can extend without forking the core

Improved knowledge extraction - higher-fidelity entity and relationship capture with fewer hallucinated graph edges

Reduced query-time compute - traversal over a structured graph instead of raw chunk retrieval

Default ontology - 11 common business concepts (People, Organizations, Products, Events, and more) that work out of the box, fully extensible with domain-specific types

LLM-agnostic - compatible with OpenAI, Anthropic, Google, Cohere, local open-source models, and 100+ others via a single interface

Multi-tenancy - isolated knowledge graphs per tenant

Onboarding UI - accelerates time-to-working-system from months to days

Predictable cost structure: ingesting a 1,000-document corpus costs approximately $5-6 in LLM tokens using GPT-4o-mini. Each retrieval query adds roughly $0.001.

Availability

GraphRAG SDK 1.0 is available now under an open-source license.

GitHub: https://github.com/FalkorDB/GraphRAG-SDK

Benchmark leaderboard: https://graphrag-bench.github.io

Dan Shalev, Marketing & Product
dan.shalev@falkordb.com
(972)51 558 4042
FalkorDB Inc., 850 New Burton Road, Suite 201 City of Dover, Country of Kent Delaware 199404

FalkorDB is a next-generation graph database designed to optimize large language models (LLMs) through its unique GraphRAG technology. Founded by ex-Redis engineers, it uses sparse matrices and GraphBLAS for efficient graph traversal and algebraic querying. FalkorDB supports vector similarity search and integrates seamlessly with AI frameworks like LangChain and LlamaIndex. Its ultra-low-latency architecture enables real-time decision-making by combining knowledge graphs with vector databases, reducing hallucinations in AI outputs. Additionally, FalkorDB offers multi-tenant capabilities and supports the Bolt protocol for easy migration from Neo4j. This makes it ideal for CTOs, Chief Architects, and AI/ML Engineers who need scalable solutions for high-performance Retrieval-Augmented Generation (RAG) applications. With its distributed architecture and in-memory processing, FalkorDB ensures fast query execution even with massive datasets, making it a preferred choice for enterprise-scale AI deployments.

This release was published on openPR.