JA EN
shiichan

GPT-5.6 Brings a 1M-Token Context Window to Amazon Bedrock!

AWS What's New ★★★★ View Markdown
#aws#ai

Hey, it's Shiichan! Today's news from AWS is about a context window that just got a lot bigger.

AWS What's New aws.amazon.com

What was announced?

According to AWS What's New, GPT-5.6 Sol, Terra, and Luna on Amazon Bedrock now support 1 million token context windows. That means you can pass a full codebase, a lengthy document, or a multi-turn agent history into a single request without splitting it up.

Why it matters

Working with long code or documents used to mean running into context limits and having to split content into chunks before feeding it to the model. Chunking brings its own risk: the model can lose track of context across chunks, or miss information that falls near a split. A window this large means many of those cases no longer need to be chunked at all.

What changes

With more room in a single request, you can now:

  • Analyze an entire repository in one pass for code review or migration work
  • Process long legal or regulatory documents end-to-end, without chunking
  • Keep a full conversation history intact across multi-step agentic workflows

Because the model reasons over more context at once, you get more accurate, coherent responses without the information loss that chunking can introduce.

Dive Deep

Here's what you should know about pricing:

  • Prompt caching with explicit cache breakpoints applies to long-context requests too
  • Repeated context is billed at a 90% discount
  • Pricing matches OpenAI's first-party rates, and usage counts toward your AWS commitments

Regional availability varies slightly by model:

  • GPT-5.6 Sol: US East (N. Virginia), US East (Ohio)
  • GPT-5.6 Terra and Luna: US East (N. Virginia), US East (Ohio), US West (Oregon)

You can get started via the Amazon Bedrock Console, or the Responses API on the bedrock-mantle endpoint.

Wrap-up

  • GPT-5.6 Sol, Terra, and Luna on Amazon Bedrock now support 1M-token context windows
  • Full codebases, long documents, and agent conversation histories can be processed without chunking
  • Repeated context is billed at a 90% discount via prompt caching
  • Sol is available in 2 regions; Terra and Luna in 3
  • Get started via the Bedrock Console or the Responses API (bedrock-mantle endpoint)

If you're reviewing huge repositories, processing long-form documents, or building agents with long-running histories, this update is exactly for you!

Related posts