Nova Lite vs Nova Pro on Amazon Bedrock

In us-east-1, Nova Lite costs $0.060 per 1M input tokens against $0.800 — a 13.3× difference. Prices, context, capabilities and region coverage below.

amazon.nova-lite-v1:0
Input / 1M
$0.060
Output / 1M
$0.240
Context
300K
Est. / month
$64.80
Nova Pro Amazon
amazon.nova-pro-v1:0
Input / 1M
$0.800
Output / 1M
$3.20
Context
300K
Est. / month
$864.00

Capabilities

Capability Nova Lite Nova Pro
Tool use
Vision
Prompt caching
Batch inference
Streaming

Region availability

Region Nova Lite Nova Pro
us-east-1 US East (N. Virginia) On-demand On-demand
us-east-2 US East (Ohio) On-demand Inference profile only
us-west-2 US West (Oregon) On-demand Inference profile only
ca-central-1 Canada (Central) Inference profile only Not available
eu-west-1 EU (Ireland) Inference profile only Inference profile only
eu-west-2 EU (London) On-demand On-demand
eu-west-3 EU (Paris) Inference profile only Inference profile only
eu-central-1 EU (Frankfurt) Inference profile only Inference profile only
eu-north-1 EU (Stockholm) On-demand Inference profile only
ap-southeast-1 Asia Pacific (Singapore) Inference profile only Inference profile only
ap-southeast-2 Asia Pacific (Sydney) On-demand On-demand
ap-northeast-1 Asia Pacific (Tokyo) On-demand Inference profile only
ap-northeast-2 Asia Pacific (Seoul) Inference profile only Inference profile only
ap-south-1 Asia Pacific (Mumbai) Inference profile only Inference profile only

13 of 16 regions support both models.

Common questions

Is Nova Lite or Nova Pro cheaper on Amazon Bedrock?
Nova Lite is cheaper on input tokens: $0.060 against $0.800 per 1M in us-east-1, a 13.3× difference. At 2,000 runs a day with 12,000 input and 1,500 output tokens each, that is $64.80 against $864.00 a month before any caching or batch discount.
Does Nova Lite or Nova Pro have the larger context window?
Both have a 300K-token context window.
Which regions support Nova Lite and Nova Pro?
Nova Lite runs in 14 of the 16 AWS regions tracked here and Nova Pro in 13. 13 regions support both, so a workload that may switch between them should target one of those.