Your Bedrock account starts with default quotas — ceilings on requests and tokens per minute for each Claude model. If your capacity math (see the quota dimensions guide) says production traffic will exceed them, you need an increase approved before launch, not after your first throttling incident. There are two distinct paths, depending on which Bedrock endpoint your traffic uses.
Path 1: bedrock-runtime, via the Service Quotas console
For the legacy InvokeModel/Converse surface, adjustable quotas go through AWS Service Quotas — the standard AWS mechanism for raising limits. The workflow, in plain terms:
Open the Service Quotas console in the region where you call Bedrock (quotas are regional — an increase in us-east-1 does nothing for eu-west-1). Find Amazon Bedrock in the service list, then locate the specific per-model quota by name — for example "On-demand InvokeModel tokens per minute" or "Cross-Region InvokeModel tokens per minute" for your Claude model. Check whether the quota is marked adjustable; if it is, submit a request with your target value. The console shows your request's status, and the same request history is where you track progress.
Two published rules shape whether you'll be approved. First, increases are not granted for models in Legacy or Deprecated lifecycle status — if you're scaling, scale on a current model. Second, priority goes to accounts whose traffic already consumes their existing allocation. A request for 10× capacity from an account using 5% of its current quota is a weak case; sustained utilization is your best supporting evidence. Ramp real traffic first, then ask.
Path 2: bedrock-mantle, via an AWS Support case
The current "Claude in Amazon Bedrock" surface plays by different rules: mantle quota increases are not processed through Service Quotas at all. Instead you open an AWS Support limit-increase case, specifying four things explicitly: the endpoint (bedrock-mantle), the region, the model, and the quota name (input or output tokens per minute). Vague cases bounce; precise ones move.
On the Anthropic side of this surface, documentation states the default quota is 2 million input tokens per minute, expandable to 4 million without additional Anthropic approval — meaning increases within that band are an AWS-process matter, not a model-provider negotiation. AWS enforces request-per-minute limits on the Bedrock side; contact AWS support for RPM adjustments too. (AWS's own mantle quota table publishes a much higher default for one model — Claude Opus 4.7 at 20M input TPM — so check the console for what your account actually has before requesting anything.)
Lead time: plan for delay, not for a date
AWS does not publish a guaranteed turnaround for Bedrock quota increases, and anyone who quotes you one is guessing. Treat approval time as uncertain and build your plan around that:
Request early. File as soon as your capacity estimate is credible — ideally when your pilot's utilization data starts supporting the ask — rather than the week before launch.
Stage the ask. A moderate increase backed by current usage, followed by a second request as traffic grows, tends to fit the "priority to accounts consuming their allocation" rule better than one speculative jump.
Have a bridge plan. Until the increase lands, you can buy headroom operationally: prompt caching (cache reads don't count against mantle input TPM), right-sizing max_tokens so mantle's admission control reserves less per request, spreading load with cross-region inference profiles, and robust backoff via retry handling.
Tracking and verifying
For runtime quotas, the Service Quotas console shows request state and, after approval, the new applied value — verify the number, don't assume it. For mantle, the Support case is your tracking thread; when it resolves, confirm the new limit is reflected in your account and run a controlled load test before declaring the launch unblocked. Keep the approved values in your capacity-planning docs so the next scaling round starts from facts.
Where to go next
Understand what you're asking for with Bedrock quota dimensions, and consider Provisioned Throughput if your need is a large, steady baseline rather than a higher on-demand ceiling.