Couchbase's Capella iQ shows why the real AI product is the inference control plane
krishnan · x · 2026-07-23
Couchbase's Capella iQ story on Amazon Bedrock shows that the real product is the inference control plane.
- The post argues that multi-model AI is not about listing many models in a slide deck, but about building the production plumbing: private networking, regional failover, timeout policies, regression tests, and model qualification.
- Couchbase initially chose Anthropic's Claude Sonnet 4.5 for Capella iQ after evaluating multiple model families through Bedrock.
- Its evaluation emphasized functional correctness, determinism, latency, and formatting consistency; the post says Sonnet 4.5 reached about 76% accuracy on an internal benchmark modeled on BIRD.
- The architecture uses EKS, a VPC interface endpoint, and Bedrock Cross-Region Inference across US regions for failover and load balancing.
The core takeaway: if a product can't switch models when quality, cost, or capacity changes, it doesn't really have a multi-model architecture yet — it has a single-model dependency with extra logos.
More from Infra
- Agentic RL setup hosts 135,000 task images and 1,000 concurrent rollouts — willccbb · 2026-07-23
- An AI engineer’s learning list covers caching, routing, RAG, and quantization — EmployerNegative5653 · 2026-07-23
- Huawei GPUs may only have a 3-year life, while NVIDIA chips keep a 5-year depreciation window — teortaxesTex · 2026-07-23
- Intel and AMD face rising China server CPU prices as lead times stretch to six months — firstadopter · 2026-07-23
- Kimi V4 is headed for native multimodality, but Moonshot says compute is the limit — teortaxesTex · 2026-07-23
- Huawei 950 supernode is described as a GB200/GB300 substitute, but capacity is still tight — teortaxesTex · 2026-07-23