Which Bedrock model is best for agentic workflows?

For agentic loops, prompt caching matters more than headline token price, because the same large prompt prefix is re-sent on every step. Filter for tool use and prompt caching, then compare on cached-input price rather than list input price.

Claude or Nova on Bedrock — which is cheaper?

Amazon Nova models are substantially cheaper per token than Anthropic Claude models of comparable tier, while Claude models generally rank higher on reasoning quality. The right answer depends on volume: at high token volume the price gap dominates, at low volume it rarely does.

Why is a model unavailable in my region?

AWS rolls Bedrock models out region by region. Many newer models are reachable from a region only through a cross-region inference profile (a us., eu. or apac. prefixed ID) rather than the bare model ID.

Which Bedrock model should I use?

Six questions. A ranked shortlist with the reasoning shown, priced against your actual volume — scored deterministically from 114 models in the daily AWS snapshot, not by asking a model to guess.

Use case1/6

What are you building?

Step 1 of 6 · Use case

Choosing a Bedrock model

Agentic workloads: caching beats sticker price

An agent re-sends its whole prompt prefix on every step. With prompt caching, those repeated tokens bill at roughly a tenth of the input rate — so a model with a higher list price and caching often costs less in practice than a cheaper model without it. The cost calculator models this directly.

Region availability is the constraint people hit first

Model availability varies sharply by region, and many models are reachable only via a cross-region inference profile. If you have a data-residency requirement, set the region constraint first — it removes more candidates than any other answer. See availability by region.

How the ranking works

Hard requirements remove candidates outright. Survivors are scored on quality, price, latency, use-case fit, context headroom, capability breadth and region coverage, re-weighted by your priority and volume. It is a plain function — same answers, same ranking, every time — which is why the result URL is shareable.