V7 reports 0.2% tool-call error rate
An OpenAI-published customer case study details how V7's knowledge-graph memory layer performs across recent models, with the vendor reporting tool-call errors on its internal benchmark falling from 2.7% with GPT-5.5 to 0.2% with GPT-5.6 Sol.
In a case study published by OpenAI, V7 — started in 2018 by Rizzoli and Edwardsson after they built a computer vision accessibility app — describes V7 Go, an agentic platform that ingests files from repositories such as SharePoint and Google Drive and resolves them into a "Context Graph" of entities, relationships and cited evidence that agents query directly, falling back to RAG when the graph lacks enough information.
The model-level numbers are the part builders will care about. V7 says GPT-5.6 Sol cut the tool-call error rate on its Context Graph benchmark from 2.7% with GPT-5.5 to 0.2%. It reports 89% accuracy for GPT-6 Astra on its hardest queries, and for GPT-5.6 Luna, 78% lower cost per document and +11.6 points of accuracy. On HERB, a benchmark for finding and connecting information spread across enterprise systems, V7 says its retrieval-only system outperformed the official baseline by 69% and reduced hallucinations on unanswerable queries by 38%.
Customer-side figures are broader: asset managers screening deals 21x faster, a financial services review cut from more than 100 hours to under 10 and saving $12,000 per task, 13.5% fewer claims-processing errors, and 50–100 step workflows completing in minutes at a claimed 99.9% accuracy.
All of it is self-reported and unaudited. The case study gives no methodology for the HERB comparison, no baseline definitions behind the accuracy deltas, and no third-party verification of the customer results.