I tried out GPT-6 Sol/Luna, which became available on Amazon Bedrock
This page has been translated by machine translation. View original
Introduction
On September 22, 2026 (UTC), OpenAI announced GPT-6 Sol/Luna, which became available on Amazon Bedrock on the same day.
In this article, we will call Sol and Luna from Bedrock in three ways — Converse, Responses API, and Chat Completions — and verify the model IDs, invocation methods, and pricing.
Overview of the GPT-6 Family
Following GPT-6 Astra, Sol and Luna have been added, bringing the lineup to three models.
| Model | Positioning | Main Use Cases |
|---|---|---|
| Astra | Flagship | Complex tasks requiring the highest quality |
| Sol | Everyday model | Complex coding, agent workflows, data analysis |
| Luna | High-efficiency model | Large-scale summarization, extraction, classification, routing |
Luna allows reasoning effort to be adjusted, enabling you to balance quality, speed, and cost on a per-request basis.
GPT-6 Sol/Luna Model List
Using list-foundation-models, you can confirm the two models, Sol and Luna.
aws bedrock list-foundation-models \
--region us-east-1 \
--by-provider OpenAI
The entries for Sol and Luna were as follows.
{
"modelArn": "arn:aws:bedrock:us-east-1::foundation-model/openai.gpt-6-sol",
"modelId": "openai.gpt-6-sol",
"modelName": "GPT-6 Sol",
"providerName": "OpenAI",
"inputModalities": ["TEXT", "IMAGE"],
"outputModalities": ["TEXT"],
"responseStreamingSupported": true,
"inferenceTypesSupported": ["INFERENCE_PROFILE"],
"modelLifecycle": {
"status": "ACTIVE",
"startOfLifeTime": "2026-09-22T17:00:00+00:00"
}
},
{
"modelArn": "arn:aws:bedrock:us-east-1::foundation-model/openai.gpt-6-luna",
"modelId": "openai.gpt-6-luna",
"modelName": "GPT-6 Luna",
"providerName": "OpenAI",
"inputModalities": ["TEXT", "IMAGE"],
"outputModalities": ["TEXT"],
"responseStreamingSupported": true,
"inferenceTypesSupported": ["INFERENCE_PROFILE"],
"modelLifecycle": {
"status": "ACTIVE",
"startOfLifeTime": "2026-09-22T17:00:00+00:00"
}
}
Since inferenceTypesSupported is INFERENCE_PROFILE only, you use an inference profile ID rather than a model ID for invocation. In this article, we verified using the Global inference profile (global.openai.gpt-6-sol / global.openai.gpt-6-luna).
Calling from Converse
aws bedrock-runtime converse \
--region us-east-1 \
--model-id global.openai.gpt-6-sol \
--messages '[{"role":"user","content":[{"text":"Say ok"}]}]'
{
"output": {
"message": {
"role": "assistant",
"content": [{"text": "ok"}]
}
},
"stopReason": "end_turn",
"usage": {
"inputTokens": 8,
"outputTokens": 5,
"totalTokens": 13,
"cacheReadInputTokens": 0,
"cacheWriteInputTokens": 0
},
"metrics": {"latencyMs": 540}
}
By replacing --model-id with global.openai.gpt-6-luna, Luna also responded in the same format (only a single text invocation was verified).
Calling from the OpenAI-Compatible Path
You can use the OpenAI-compatible API under /openai/v1/ of the bedrock-runtime endpoint.
We called the Responses API with Sol. Temporary credentials are used for SigV4 authentication. --aws-sigv4 requires curl 7.75 or later.
cat gpt6-sol-responses-body.json
# {"model":"global.openai.gpt-6-sol","input":"Say ok","store":false}
curl --silent --show-error \
--aws-sigv4 'aws:amz:us-east-1:bedrock' \
--user "$AWS_ACCESS_KEY_ID:$AWS_SECRET_ACCESS_KEY" \
-H "x-amz-security-token: $AWS_SESSION_TOKEN" \
-H 'Content-Type: application/json' \
--data-binary @gpt6-sol-responses-body.json \
-w '\nHTTP %{http_code}\n' \
https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1/responses
HTTP 200 was returned, the status in the response body was completed, and model was global.openai.gpt-6-sol.
Chat Completions was called with Luna.
curl --silent --show-error \
--aws-sigv4 'aws:amz:us-east-1:bedrock' \
--user "$AWS_ACCESS_KEY_ID:$AWS_SECRET_ACCESS_KEY" \
-H "x-amz-security-token: $AWS_SESSION_TOKEN" \
-H 'Content-Type: application/json' \
-d '{"model":"global.openai.gpt-6-luna","messages":[{"role":"user","content":"Say ok"}],"max_completion_tokens":16}' \
-w '\nHTTP %{http_code}\n' \
https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1/chat/completions
HTTP 200 was returned, finish_reason was stop, and model was global.openai.gpt-6-luna.
Pricing
These are the prices from OpenAI direct sales (as of 2026-09-22, based on OpenAI's announcement).
| Model | Input (/1M tokens) | Output (/1M tokens) |
|---|---|---|
| GPT-6 Sol | $2.00 | $10.00 |
| GPT-6 Luna | $0.10 | $0.50 |
Prices have been reduced compared to the equivalent models in the GPT-5.6 series.
At the time of writing, we were unable to confirm the pricing for GPT-6 Sol/Luna on the Bedrock pricing page. When using the Global inference profile, OpenAI's pricing is expected to apply (with GPT-5.6, the Global inference profile pricing was the same as OpenAI direct sales).
Summary
We confirmed invocation from the bedrock-runtime endpoint in three ways: Converse, Responses API, and Chat Completions. The inference profile IDs are global.openai.gpt-6-sol / global.openai.gpt-6-luna.
Pricing on Bedrock could not be confirmed at the time of writing, but based on the precedent of GPT-5.6, it is expected that the same price as OpenAI will apply when using the Global inference profile.
Since all models have been reduced in price compared to GPT-5.6, those with GPT-5.6 workloads may find it worthwhile to consider migrating to GPT-6.
GPT-6 is provided via the bedrock-runtime endpoint and already supports a 1M token context. Even for workloads that have traditionally used Claude-based models, GPT is now a viable option, so please give it a try.
References
