OpenAI GPT-5.6 Sol / Terra / Luna Are Now Supported on Amazon Bedrock, So I Tried Them Out

Introduction

The GPT-5.6 family that OpenAI announced on July 9, 2026 became GA on Amazon Bedrock on July 13. It became available on Bedrock just a few days after its release.

https://aws.amazon.com/blogs/machine-learning/openai-gpt-5-6-sol-terra-and-luna-are-now-generally-available-on-amazon-bedrock/

https://openai.com/ja-JP/index/gpt-5-6/

The naming scheme has changed with GPT-5.6. The number 5.6 represents the generation, and Sol / Terra / Luna represent tiers that evolve independently. The flagship Sol does full-power reasoning, Terra is the balanced everyday model, and Luna is the lightweight, low-cost tier.

Here is a comparison of the GPT-5.6 family and recent models, confirmed as of July 14, 2026. Prices are unit prices per 1M tokens (USD).

※ The context length is the value listed in the Amazon Bedrock model card. OpenAI's official documentation states a 1,050K context window, but on Bedrock it is 272K.

In Bedrock pricing, Sol is cheaper than GPT-5.5 for both Input and Output, and Terra is cheaper than GPT-5.4. Luna was newly introduced as the lowest-priced tier in the family.

The authentication and setup steps are the same as in the previous article, so please refer to it.

https://dev.classmethod.jp/articles/bedrock-openai-gpt55-gpt54-ga-responses-api/

Model IDs, pricing, and region support were confirmed in the following official documentation.

• Model card
• Pricing
• Region support

Verification Details

Test Environment

The GPT-5.6 family is exclusive to the bedrock-mantle endpoint, not the bedrock-runtime endpoint. It does not support the Invoke API / Converse API, and is called through the OpenAI-compatible Responses API.

Here is the client initialization.

from openai import OpenAI
from aws_bedrock_token_generator import BedrockTokenGenerator

REGION = "us-east-1"
BASE_URL = f"https://bedrock-mantle.{REGION}.api.aws/openai/v1"

generator = BedrockTokenGenerator()
token = generator.get_token(credentials, REGION)

client = OpenAI(base_url=BASE_URL, api_key=token)

Non-Streaming Calls

I sent the same prompt to the three models in non-streaming mode and confirmed that they can be called successfully. reasoning.effort is set to low for all of them.

response = client.responses.create(
    model="openai.gpt-5.6-sol",
    input="Amazon Bedrockとは何ですか？3文で簡潔に説明してください。",
    reasoning={"effort": "low"},
)

Here is an example response from Sol.

Amazon Bedrock is a fully managed service provided by AWS for developing generative AI applications. It lets you use foundation models offered by multiple leading companies via API, and select, customize, and evaluate models. While keeping infrastructure management to a minimum, you can securely build chatbots, text generation, search assistance, and more.

All three models responded normally. Each is a single measurement and should be treated as a reference value.

The initial latency for Terra and Luna was large at about 75 seconds, but since it dropped significantly from the second call onward, it is likely due to a cold start. On the second call, Terra dropped to 1.12s and Luna to 0.57s. Sol was 2.22s from the first call, so it may already have been warm.

For a prompt requesting a simple explanation, reasoning_tokens was 0 for all three models.

Sol:
ResponseUsage(input_tokens=23, input_tokens_details=InputTokensDetails(cached_tokens=0, cache_write_tokens=0), output_tokens=102, output_tokens_details=OutputTokensDetails(reasoning_tokens=0), total_tokens=125)

Terra:
ResponseUsage(input_tokens=23, input_tokens_details=InputTokensDetails(cached_tokens=0, cache_write_tokens=21), output_tokens=107, output_tokens_details=OutputTokensDetails(reasoning_tokens=0), total_tokens=130)

Luna:
ResponseUsage(input_tokens=23, input_tokens_details=InputTokensDetails(cached_tokens=0, cache_write_tokens=21), output_tokens=115, output_tokens_details=OutputTokensDetails(reasoning_tokens=0), total_tokens=138)

Streaming Calls

I called the same prompt with stream=True and confirmed the streaming response.

stream = client.responses.create(
    model="openai.gpt-5.6-sol",
    input="Amazon Bedrockとは何ですか？3文で簡潔に説明してください。",
    reasoning={"effort": "low"},
    stream=True,
)

These are the results measured in a warm state. TTFT is the time until the first response.output_text.delta event is received.

The event structure was the same across the three models. It starts with response.created, response.output_text.delta is sent repeatedly, and it ends with response.completed.

Checking reasoning.effort

I called Sol while varying reasoning.effort and checked how reasoning_tokens changed. The prompt is "Prove that there are infinitely many prime numbers." I did not evaluate accuracy; I only looked at whether it works and at the token counts.

response = client.responses.create(
    model="openai.gpt-5.6-sol",
    input="素数が無限に存在することを証明してください。",
    reasoning={"effort": "max"},
)

With the simple explanation prompt in non-streaming mode, reasoning_tokens=0, but with the prompt requesting a proof, reasoning tokens were returned. From low to high, reasoning_tokens stayed roughly flat at 33–37, and no clear increase due to effort was observed. Latency was also inverted, from low 3.34s to high 2.50s, and since each is a single measurement, this is within the range of measurement error. Only at max did reasoning_tokens clearly increase to 104.

Specifying reasoning.effort="ultra" returned an error.

Invalid value: 'ultra'. Supported values are: 'none', 'minimal', 'low', 'medium', 'high', 'xhigh', and 'max'.

The values that can be specified for reasoning.effort on Bedrock Mantle are seven levels: none / minimal / low / medium / high / xhigh / max. ultra was not accepted at the time of this verification.

Confirming Chat Completions API Is Not Supported

At least at the time of this verification, Bedrock Mantle does not support the Chat Completions API, and calling it returned the following error.

client.chat.completions.create(
    model="openai.gpt-5.6-sol",
    messages=[{"role": "user", "content": "Hello"}],
)

400 - {'error': {'code': 'validation_error', 'message': "The model 'openai.gpt-5.6-sol' does not support the '/v1/chat/completions' API", 'param': None, 'type': 'invalid_request_error'}}

Summary

I called GPT-5.6 Sol / Terra / Luna, which became GA on Amazon Bedrock, from the Responses API on the Bedrock Mantle endpoint. All three models responded normally in both non-streaming and streaming modes, confirming that the GPT-5.6 family can be chosen according to use case on Bedrock as well.

In Bedrock pricing, Sol is cheaper than GPT-5.5 and Terra is cheaper than GPT-5.4. Luna was added as a low-cost tier at $1/$6, and it looks like a good first candidate to try for lightweight tasks such as summarization and classification.

At the time of this verification, the Service Tier was Standard only, and Priority / Flex / Reserved and cross-region inference were not supported (In-Region only). Image input is supported, but it was not verified this time.

As a point of caution, the model card states that traffic data flagged by the classifier through abuse detection is retained for up to 30 days. Since this is an exception to the usual Bedrock zero-data-retention policy, it would be wise to check the model card before actual use.