Model & Region Notes

Community notes on what is actually available where, and what changed.

⚠️ This page is deliberately not in the repository. Model and region availability changes monthly; a versioned file would be wrong within a quarter and quietly mislead people. Here it can be corrected the moment somebody notices.

AWS is authoritative, not this page. Always verify with the commands below before relying on anything written here.

---

Verify it yourself — always start here

# models you can call in a region
aws bedrock list-foundation-models --region us-east-1 \
  --query 'modelSummaries[].modelId' --output table

# inference profile IDs — these are what you pass as modelId
aws bedrock list-inference-profiles --region us-east-1 \
  --query 'inferenceProfileSummaries[].[inferenceProfileId,status]' --output table

An empty list is a permissions state, not an outage — model access has not been granted in that region.

---

The rules that do not change

These are stable even as the model list churns, which is why they are worth writing down:

1. Access is per model, per region, on request. Not automatic, and approval is not always instant.
2. Many models require an inference profile ID, not the bare model ID. The profile carries a geography prefix — us., eu., and others.
3. A geographic profile routes across several regions, and you need model access in every destination — not just the one you call from. This is why access failures can look intermittent.
4. Feature availability lags model availability. A region having a model does not mean it has Knowledge Bases, Guardrails or AgentCore.
5. us-east-1 has the widest support. This curriculum is developed against it. Elsewhere, verify before starting a module.

---

Region notes

Region	Notes	Last checked	By
us-east-1	Curriculum baseline. Widest model and feature coverage	—	maintainer
us-west-2	Generally good coverage; verify AgentCore before Module 11	—	—
eu-*	Model set differs; eu. profile prefix. Verify per module	—	—
ap-*	Verify before starting. Feature gaps more likely	—	—

Add a row when you verify one. Region, what worked, what did not, and the date — the date is the important column.

---

Feature availability, by module

Before starting a module in a non-baseline region, check the thing it actually needs:

Module	Needs	Check
02 Bedrock	Converse API, a chat model	list-foundation-models
02 / 04	Knowledge Bases, Guardrails	Console — feature presence varies
03 Bedrock Agents	Bedrock Agents + Lambda	Console
10 Retrieval	OpenSearch Serverless, an embedding model	aws opensearchserverless list-collections
11 AgentCore	AgentCore — not available everywhere	Console

---

Model choice, when the list changes under you

The model selection cheat sheet is deliberately written to survive this: choose on four dimensions (capability, latency, context, cost ceiling), record what you rejected and why, and re-decide on a stated trigger.

The one number worth carrying between models: cost per resolved task, not cost per token. A cheaper model needing two extra turns is not cheaper.

---

Changes worth recording

When something changes in a way that breaks material here, note it — and open an issue so the repo can be fixed.

Date	What changed	Affects	Reported by
—	—	—	—

AWS behaviour changing is the single highest-value thing to report here. Nothing degrades teaching material faster.