OpenAI's GPT-5.6 Sol, Terra, and Luna become generally available on Amazon Bedrock

Amazon Web Services announced on July 24, 2026 that OpenAI's GPT-5.6 model family — Sol, Terra, and Luna — is now generally available on Amazon Bedrock, giving AWS customers direct access to OpenAI's current flagship models alongside Bedrock's existing model catalog.

What's new

AWS states plainly: "OpenAI GPT-5.6 Sol, Terra, and Luna are now generally available on Amazon Bedrock." The models are accessed through the OpenAI Responses API via a bedrock-mantle endpoint, using standard OpenAI SDKs rather than requiring customers to rewrite integrations in Bedrock-native call formats.

On capabilities, the Bedrock-hosted models carry a 272,000-token context window, accept text and image input with text output, support configurable reasoning effort from "none" up to "max," and include tool calling plus both implicit and explicit prompt caching. Pricing matches OpenAI's own first-party rates, with usage counted against a customer's existing AWS spend commitments — and cached input tokens are discounted 90% versus uncached input, a meaningful cost lever for high-repetition agentic workloads.

Availability is currently US-only and split by model: GPT-5.6 Sol is live in US East (N. Virginia) and US East (Ohio), while Terra and Luna add US West (Oregon) to that list.

Context

This is the second frontier-model cross-listing of its kind in recent weeks — xAI's Grok 4.3 became generally available on Amazon Bedrock shortly before this announcement, and Anthropic's Claude models have long been available on Bedrock as one of the platform's original model partners. AWS has been steadily expanding Bedrock's role as a neutral hosting layer where competing labs' models sit side by side, rather than pushing customers exclusively toward AWS's own in-house models.

For OpenAI, it extends GPT-5.6's distribution beyond its own API and Microsoft's Azure OpenAI Service into a second major hyperscaler's marketplace, following a pattern where frontier labs increasingly ship simultaneously across multiple clouds rather than picking one exclusive distribution partner.

Why it matters

Multi-cloud availability lowers switching costs for enterprises that have already standardized their infrastructure, compliance, and billing around a particular cloud provider but want access to a specific model family for a given workload. An AWS shop that wants GPT-5.6's reasoning-effort controls no longer needs a separate OpenAI billing relationship or a second vendor security review to get it.

It also reflects how commoditized frontier-model access is becoming at the infrastructure layer: rather than compete purely on exclusive access, cloud providers are competing on developer experience, pricing mechanics like cache discounts, and breadth of catalog — with the underlying frontier models from OpenAI, Anthropic, and xAI increasingly available as interchangeable options within the same console.