AWS brings GPT-6 Sol and Luna to Bedrock and launches Strands Harness

AWS moved quickly to put the week's new frontier models into enterprise hands, making OpenAI's GPT-6 Sol and GPT-6 Luna generally available on Amazon Bedrock. The models support extensive context windows — up to 1 million input tokens — and are pitched for deeper reasoning, professional-quality writing, and advanced computer use, all within Bedrock's secure, governed, and auditable inference environment. This lets AWS customers deploy the new cheaper GPT-6 tiers without leaving their existing governance and compliance boundaries.

Separately, AWS launched Strands Harness, a fully-assembled, state-of-the-art agent harness that extends the open-source Strands Agents SDK first released in May 2025. VP Swami Sivasubramanian says it delivers frontier performance at 28% lower token cost than comparable harnesses while matching or beating them on accuracy, with much of the efficiency coming from smart defaults like prompt caching and context management that truncates bulky tool results. It runs locally or deploys to any provider, supports models from Bedrock, Anthropic, OpenAI, and Google plus local Ollama models, and is open-source under Apache 2.0.

Together the two releases reinforce AWS's neutral-platform strategy: rather than betting on a single model, AWS offers the widest menu (Anthropic, OpenAI, Amazon Nova, and more) plus the tooling to orchestrate agents cost-effectively across them. The 28% cost claim also fits the week's dominant theme of driving down inference economics.

The caveat is vendor-reported benchmarking — the 28% figure and accuracy claims come from AWS. Still, an open-source, Apache-2.0 harness with cross-provider support is directly useful to builders. Watch adoption relative to competing agent frameworks like Microsoft's Agent Framework.