Pinecone Nexus cuts agent task cost 77%

Pinecone made Nexus generally available after its knowledge layer held GPT-5.5 accuracy while cutting cost per task 77% on 97 τ-Knowledge customer-service tasks. Pinecone’s GA benchmark report says GPT-5.2 with Nexus also gained 12 percentage points of accuracy, cut cost 80%, and reduced model calls from 81.7 to 42.6 per task. Builders should file those vendor-run results beside the case for routing models and tracing agent costs: retrieval architecture can erase more spend than swapping one frontier model for another, but the claim still needs replication on a buyer’s own corpus.