Amazon Bedrock adds OpenAI GPT-5.6 models with cross-Region inference

Amazon Bedrock expanded its model catalog to include OpenAI's GPT-5.6 family — Sol, Terra and Luna — accessible on the standard bedrock-runtime endpoint and exposed through Bedrock's Responses, Converse and Chat Completions APIs. That triple-API support matters because it lets existing Bedrock customers swap OpenAI models into pipelines already built around Converse or the newer Responses interface without rewriting integration code.

The headline capability is cross-Region inference for OpenAI models. Using Global and Geo routing, Bedrock can dynamically dispatch inference requests across AWS regions to raise throughput and lower cost, while Geo routing keeps traffic within a geographic boundary for data-residency and regulatory compliance. For enterprises running high-volume or latency-sensitive agentic workloads, this addresses two perennial pain points at once: capacity ceilings during peak demand and jurisdictional constraints on where data may be processed.

The move deepens the increasingly complex AWS-OpenAI-Anthropic triangle. AWS remains Anthropic's primary cloud partner, yet it is now merchandising OpenAI's frontier models directly on Bedrock — a pragmatic, model-neutral marketplace strategy that mirrors how AWS also hosts NVIDIA's Nemotron 3.5 Lightning via SageMaker JumpStart and OpenAI's 'Daybreak' cybersecurity initiative. AWS's Q2 backdrop underscores the stakes: net sales up 36.7% year-on-year to $42.2B, its fastest growth in 18 quarters.

For customers the practical questions are pricing parity versus OpenAI's own API, whether the Global routing introduces latency variance, and how model-version updates propagate through Bedrock's managed layer. But strategically, the message is clear: AWS wants to be the neutral inference substrate for every frontier model, not just its Anthropic investment.