OpenAI GPT-5.6 models arrive on Amazon Bedrock with explicit prompt caching

The Bedrock availability makes GPT-5.6 Sol, Terra, and Luna first-class citizens on AWS's managed inference platform, and the headline feature is explicit prompt caching — a mechanism, documented jointly by AWS and OpenAI, that lets developers designate exactly which prompt segments are cached and reused across calls. For workloads with large, stable system prompts or shared context, that materially reduces per-request cost when migrating existing GPT workloads to Bedrock.

The launch coincided with OpenAI's 80%/20% price cuts to Luna and Terra, which Bedrock mirrored the same day, compounding the cost savings for AWS customers. AWS also highlighted a customer proof point: AUMOVIO used multi-agent AI on Bedrock to catch automotive embedded-code defects at scale, illustrating the agentic use cases the cheaper, cacheable models target.

Competitively, this deepens Bedrock's multi-model strategy — OpenAI, Anthropic (Claude Opus 5), DeepSeek, Meta and others under one roof — which AWS leaned on heavily in its 37%-growth quarter. Offering OpenAI models on Bedrock also blurs the old Azure-OpenAI exclusivity narrative. What to watch: how explicit caching compares to Anthropic's own prompt caching on Bedrock in real cost savings, and whether enterprises consolidate multi-vendor agent stacks on Bedrock to exploit unified caching and routing.