Amazon Nova 2 Lite
Amazon's low-cost reasoning model on Bedrock, with a 1M-token context and built-in tools
Context window
1M
Input / 1M tokens
$0.30
Output / 1M tokens
$2.50
Provider
Amazon Web Services
Amazon Bedrock on-demand, per 1M tokens: $0.30 input and $2.50 output through global cross-Region inference, or $0.33 and $2.75 through geo (US, EU, Japan) inference. Cached reads cost $0.075 (global) or $0.0825 (geo). Batch and Flex are half price, and web grounding costs $0.03 per request. Prices from the AWS price list of November 11, 2026. · Data verified 2026-11-11
Amazon Nova 2 Lite (December 2, 2025) is Amazon's fast, cost-efficient reasoning model on Amazon Bedrock for everyday workloads such as automation, document processing and customer support. It accepts text, image, video and document input with a 1M-token context and up to 64K output tokens. Extended thinking is off by default and can be turned on at low, medium or high effort. It has two built-in tools, web grounding with citations and a code interpreter, and also supports client-side tool calling and remote MCP tools. Its knowledge cutoff is October 2025, and AWS lists it as Active with no end-of-life date set.
How to access it
Amazon Bedrock (bedrock-runtime) through the Converse or Invoke API. The base model ID is amazon.nova-2-lite-v1:0, called through cross-Region inference profiles: global.amazon.nova-2-lite-v1:0, or the us., eu. and jp. prefixes for geo routing. Authenticate with AWS credentials or a Bedrock API key using boto3. Also on OpenRouter as amazon/nova-2-lite-v1.
Strengths
- ✓Low price: $0.30 input / $2.50 output per 1M tokens through global inference
- ✓1M-token context with text, image, video and document input
- ✓Adjustable extended thinking (low, medium, high)
- ✓Built-in web grounding and code interpreter on Bedrock
- ✓Fine-tuning on Bedrock and SageMaker
Best for developers who...
When to choose it (and when not to)
Reach for Amazon Nova 2 Lite when...
- →You build on AWS and want a cheap Bedrock-native model with reasoning
- →You need a long multimodal context (1M tokens) that includes video
- →You want built-in web grounding or a code interpreter without running your own tool servers
Look elsewhere if...
- ✕You need top coding or reasoning results
- ✕You need speech input or output (use Nova 2 Sonic)
- ✕You are not on AWS and do not want Bedrock as a dependency
How to use it
- ›Leave extended thinking off for simple calls and enable it with reasoningConfig for multi-step tasks
- ›Use the global inference profile for the lowest per-token price
- ›Cache long repeated prefixes; Nova models cache up to 20K tokens
Quickstart
Pythonimport boto3
client = boto3.client("bedrock-runtime", region_name="us-east-1")
resp = client.converse(
modelId="global.amazon.nova-2-lite-v1:0",
messages=[{"role": "user", "content": [{"text": "Summarize the key points of this support ticket: ..."}]}],
)
for block in resp["output"]["message"]["content"]:
if "text" in block:
print(block["text"])Requires boto3 and AWS credentials (or a Bedrock API key in AWS_BEARER_TOKEN_BEDROCK). Uses the global inference profile from AWS's launch post; add additionalModelRequestFields={"reasoningConfig": {"type": "enabled", "maxReasoningEffort": "low"}} to turn on extended thinking.
API model id: amazon.nova-2-lite-v1:0
Compare Amazon Nova 2 Lite
Compare Amazon Nova 2 Lite with any other model
Build a comparison →All model comparisons →