AWS Bedrock adds Claude Opus 4.7 and GPT-5.5

AWS shipped Claude Opus 4.7 in Amazon Bedrock and brought GPT-5.5, Codex, and Managed Agents to Bedrock in limited preview [AWS Bedrock blog].

── What shipped ──

• Claude Opus 4.7 lands in Bedrock with the next-generation inference engine purpose-built for generative AI workloads.
• GPT-5.5 + Codex + Managed Agents in preview for select customers.
• AgentCore additions: a managed harness (preview), the AgentCore CLI, and AgentCore skills for coding assistants. The harness lets you define an agent with model + system prompt + tools and run immediately, with no orchestration code [AWS weekly roundup].
• AgentCore Evaluations ships pre-built metrics for task success, response quality, and tool accuracy.
• 80% of 600 agents in a recent AWS hackathon were built on AgentCore — a self-reported but useful adoption signal.

── Why it matters ──

AWS Bedrock is now the most comprehensively multi-vendor AI plane in the major-cloud market. Anthropic Claude, OpenAI GPT, Meta Llama, and Mistral all run inside Bedrock under AWS's IAM, networking, and compliance boundary. That is a procurement profile no other cloud can match for enterprise buyers worried about model-vendor lock-in.

The AgentCore additions are AWS's response to the agent-runtime competition. Cloudflare ships Sandboxes, Vercel ships AI SDK, GCP ships Vertex AI Agent Builder. AWS is pitching a heavier-weight alternative: production-ready agent infrastructure with first-class evaluation built in.

For enterprise architects:

• Lock-in risk is now lowest at AWS for AI workloads. The argument that "GCP has Anthropic, Azure has OpenAI" is over — Bedrock has both.
• AgentCore evaluations are the real differentiator. Production agent deployment without rigorous eval is malpractice; AWS shipping eval primitives in the same SDK as the agent runtime is the right shape.

── Editor's take ──

The Bedrock catalogue is now the procurement-friendly choice for enterprises that want frontier models without picking a horse. The trade-off is that Bedrock's API surface lags the native vendor APIs by one to four weeks on new model features. For most enterprise workloads, that lag is acceptable. For frontier development, go direct.