Nova 2 Lite vs MiniMax M2.1 on Amazon Bedrock

In us-east-1, MiniMax M2.1 costs $0.300 per 1M input tokens against $0.300. Prices, context, capabilities and region coverage below.

amazon.nova-2-lite-v1:0
Input / 1M
$0.300
Output / 1M
$2.50
Context
Est. / month
$441.00
minimax.minimax-m2.1
Input / 1M
$0.300
Output / 1M
$1.20
Context
Est. / month
$324.00

Capabilities

Capability Nova 2 Lite MiniMax M2.1
Tool use
Vision
Prompt caching
Batch inference
Streaming

Region availability

Region Nova 2 Lite MiniMax M2.1
us-east-1 US East (N. Virginia) Inference profile only On-demand
us-east-2 US East (Ohio) Inference profile only On-demand
us-west-2 US West (Oregon) Inference profile only On-demand
ca-central-1 Canada (Central) Inference profile only Not available
eu-west-1 EU (Ireland) Inference profile only On-demand
eu-west-2 EU (London) Inference profile only On-demand
eu-west-3 EU (Paris) Inference profile only Not available
eu-central-1 EU (Frankfurt) Inference profile only On-demand
eu-north-1 EU (Stockholm) Inference profile only On-demand
ap-southeast-1 Asia Pacific (Singapore) Inference profile only Not available
ap-southeast-2 Asia Pacific (Sydney) Inference profile only On-demand
ap-northeast-1 Asia Pacific (Tokyo) Inference profile only On-demand
ap-northeast-2 Asia Pacific (Seoul) Inference profile only Not available
ap-south-1 Asia Pacific (Mumbai) Inference profile only On-demand
sa-east-1 South America (Sao Paulo) Not available On-demand

10 of 16 regions support both models.

Common questions

Is Nova 2 Lite or MiniMax M2.1 cheaper on Amazon Bedrock?
MiniMax M2.1 is cheaper on input tokens: $0.300 against $0.300 per 1M in us-east-1. At 2,000 runs a day with 12,000 input and 1,500 output tokens each, that is $324.00 against $441.00 a month before any caching or batch discount.
Does Nova 2 Lite or MiniMax M2.1 have the larger context window?
Both have a —-token context window.
Which regions support Nova 2 Lite and MiniMax M2.1?
Nova 2 Lite runs in 14 of the 16 AWS regions tracked here and MiniMax M2.1 in 11. 10 regions support both, so a workload that may switch between them should target one of those.