All insights

Model APIs

Kimi K2.5 API: Availability and a First Request

See a documented Kimi K2.5 route through Amazon Bedrock, plus a Python first-call example and the Region, credentials, and permission checks to make before running it.

3 min readTalk to our API team
Illustration for Kimi k2 5 api

The short answer

“Kimi K2 5 API” most likely refers to Kimi K2.5. Amazon Bedrock documents the model ID moonshotai.kimi-k2.5, lists Converse as a supported API, and provides a Bedrock Runtime endpoint pattern. The example below uses Amazon Bedrock, not Moonshot’s direct API. AWS’s Kimi K2.5 model card

Moonshot announced Kimi K2.5 on its own API platform on February 4, 2026. That announcement establishes the launch, but it is not a complete, current integration guide. Moonshot’s announcement

Together AI’s model page says Kimi K2.5 is unavailable on its Serverless API. That specific limitation does not establish whether another Together deployment option supports the model. Together AI’s model page

Make a first request through Amazon Bedrock

You need Boto3, AWS credentials, and an identity allowed to invoke the model. To configure credentials interactively with the AWS CLI, run aws configure; Boto3 also supports other credential sources. Converse requires the bedrock:InvokeModel permission. Boto3 credentials guide · Boto3 Converse reference

Install Boto3 if needed:

python -m pip install boto3

AWS’s Kimi K2.5 model card shows us-east-1 (N. Virginia) as an example when describing the regional Bedrock Runtime endpoint. Model availability can depend on Region and account access, so confirm Kimi K2.5 is available to your account in the Region you select before running the example. AWS’s Kimi K2.5 model card · AWS regional availability guidance

import boto3

client = boto3.client("bedrock-runtime", region_name="us-east-1")

response = client.converse(
    modelId="moonshotai.kimi-k2.5",
    messages=[
        {
            "role": "user",
            "content": [{"text": "Reply with: Kimi API connected"}],
        }
    ],
    inferenceConfig={"maxTokens": 128},
)

content_blocks = response["output"]["message"]["content"]
text = "\n".join(block["text"] for block in content_blocks if "text" in block)
print(text)

The request uses the model ID and Converse operation documented by AWS. The response extraction reads text blocks from the returned message content. A successful call prints the model’s response; an access or availability error means you should check your credentials, bedrock:InvokeModel permission, selected Region, and account access. AWS Kimi K2.5 model card · AWS Converse reference

Direct Moonshot API, Together, and pricing

If you specifically need Moonshot’s own API rather than Bedrock, use Moonshot’s current platform documentation to confirm the endpoint, authentication, and model ID for your account. The February 2026 announcement confirms the API launch, but does not provide a complete first-call recipe. Moonshot’s announcement

For Together AI, the documented limitation is specifically that Kimi K2.5 is not available on Serverless. The page does not establish Dedicated availability. Together AI’s model page

The sources reviewed here do not establish a single current Kimi K2.5 price across providers. A secondary article gives conflicting output-token prices, and AWS directs readers to its pricing information rather than listing a Kimi K2.5 rate on the model card. Check the price for the provider and channel you plan to use; do not transfer a price from another host or Kimi model.

Production checks before rollout

  • Keep provider-specific model IDs separate. moonshotai.kimi-k2.5 is the identifier AWS documents for Bedrock; do not assume it is also the identifier for Moonshot’s direct API or another host. AWS Kimi K2.5 model card
  • Confirm Region availability and permissions. The model card documents the Bedrock Runtime endpoint pattern and shows an us-east-1 endpoint example. Converse requires bedrock:InvokeModel; check model availability for your account and chosen Region. AWS Kimi K2.5 model card · AWS Converse reference
  • Check price and usage limits for the selected route. AWS lists a 256K-token context window and a 16K maximum output on its model card, but those specifications do not replace checking current API quotas and request constraints. AWS Kimi K2.5 model card

Sources

Kimi K2.5 — Amazon Bedrock

converse — Boto3 documentation

Credentials — Boto3 documentation

Regional availability by models — Amazon Bedrock

Kimi K2.5 API is now available — Kimi Forum

Kimi K2.5 — Together AI

Contact us