OpenAI Upgrades Its Agents SDK With Sandboxing and a New Model Harness

OpenAI has released a major update to its Agents SDK, adding two key capabilities: native sandboxing and an in-distribution model harness. Sandboxing allows agents to operate in isolated, controlled workspaces, reducing risk in production environments. The harness brings Codex-style scaffolding — including instructions, tools, approvals, tracing, handoffs, and resume bookkeeping — to the broader SDK. The harness launches in Python first, with TypeScript support planned later. The SDK remains provider-agnostic, supporting 100+ LLMs via the Chat Completions API. Additionally, OpenAI is deprecating the Assistants API with a mid-2026 sunset, urging teams to begin migrating to the Responses API now.