---
type: "post"
stable_id: "post:capella-iq-multimodel-production-loop"
slug: "capella-iq-multimodel-production-loop"
title: "Capella iQ Treats Model Choice As Configuration Backed By A Continuous Benchmark Loop"
description: "Capella iQ separates tenant and provider configuration from application logic, uses private Bedrock connectivity and cross-region inference, and continuously benchmarks models before promotion."
retrieval_nugget: "The production path runs from EKS through a VPC endpoint to Bedrock, with namespace-level model settings and region failover. The reported 76 percent accuracy is an internal Couchbase/AWS evaluation, not an external universal score."
published_at: "2026-07-20"
updated_at: "2026-08-06"
record_date: "2026-07-20"
date_kind: "published_at"
topics: ["multi-model","model-routing","resilience","amazon-bedrock","benchmarks"]
entities: ["Couchbase","Capella iQ","Amazon Bedrock"]
source_urls: ["https://aws.amazon.com/blogs/machine-learning/how-couchbase-built-a-multi-model-ai-architecture-for-capella-iq-with-amazon-bedrock"]
source_format: "article"
editorial_timing: {"lane":"regular_hourly","scheduled_at":"2026-08-08T09:00:00+03:00","real_news_delta":"owner-approved primary-source mechanism or merged analysis"}
visual_decision: {"status":"included","reason":"the central mechanism is a flow, loop, architecture, decision, or state transition that benefits from a diagram","reviewed_by":"codex"}
schema_version: "newruntime-agent-readable-v0.2"
status: "published"
visuals: [{"role":"hero","src":"/images/drip/capella-iq-multimodel-production-loop/capella-iq-multimodel-production-loop.webp","alt":"A whiteboard production architecture showing tenant configuration, private model invocation, cross-region inference, and a continuous benchmark-and-promotion loop.","caption":"New Runtime synthesis from How Couchbase built a multi-model AI architecture for Capella iQ with Amazon Bedrock."}]
routes: {"html":"https://newruntime.com/posts/capella-iq-multimodel-production-loop/","markdown":"https://newruntime.com/posts/capella-iq-multimodel-production-loop.md","json":"https://newruntime.com/posts/capella-iq-multimodel-production-loop.json"}
---

# Capella iQ Treats Model Choice As Configuration Backed By A Continuous Benchmark Loop

## Retrieval answer

The production path runs from EKS through a VPC endpoint to Bedrock, with namespace-level model settings and region failover. The reported 76 percent accuracy is an internal Couchbase/AWS evaluation, not an external universal score.

Couchbase's Capella iQ architecture treats model selection as a configuration decision rather than a code rewrite. API services assemble prompts and context; namespace configuration carries tenant preferences and provider overrides; a private VPC endpoint invokes Amazon Bedrock; cross-region inference distributes traffic and handles regional degradation.

Provider abstraction is backed by evaluation. Couchbase built a suite across SQL++ generation, index recommendations, query explanations, insights, and multi-turn conversation, scoring correctness, determinism, latency, and formatting. Claude Sonnet 4.5 reached about 76 percent accuracy in that internal evaluation; the number should remain a Couchbase/AWS claim tied to this workload.

The durable pattern is a continuous qualification loop. New models enter standardized tests, normalization, controlled traffic, failover exercises, and observability before promotion. Multi-model readiness is not a single gateway integration; it is ongoing investment in prompt contracts, benchmarks, routing, resilience, and rollback.