OpenAI Launches Responses API & Agents SDK: Powering Enterprise AI Automation

• Responses API Launch: OpenAI released the Responses API, a toolkit for building AI agents that can search the web, analyze documents, and control computers. It integrates GPT-4o and GPT-4o mini models and includes:
  • Web Search Tool: Based on ChatGPT’s existing search capabilities, providing real-time data with citations.
  • Document Search Feature: Helps sift through large files (e.g., legal cases, FAQs).
  • Computer-Use Functionality: Leverages OpenAI’s Operator model to automate tasks on a user’s device.
• Agents SDK Introduction: The Agents SDK allows developers to coordinate workflows across multiple AI agents, enabling them to collaborate on complex objectives. Nikunj Handa, OpenAI API product manager, describes it as “orchestrating atomic units of AI tools.”
• Responses API Enhancements:
  • GPT-4o Search Models: Introduces gpt-4o-search-preview (90% accuracy) and gpt-4o-mini-search-preview (88% accuracy) on the SimpleQA benchmark, priced at $30 and $25 per 1,000 queries, respectively.
  • Computer-Using Agent (CUA): Achieves 38.1% success on OSWorld (operating system tasks), 58.1% on WebArena (browser automation), and 87% on WebVoyager (web interactions). Pricing: $3 per million input tokens, $12 per million output tokens.
  • File Search Utility: Costs $2.50 per 1,000 queries + $0.10/GB/day for storage (first GB free).
• Agents SDK Upgrades:
  • **Open-Source Framework:** Inspired by Pydantic, Griffe, and MkDocs, featuring Python support (Node.js coming soon).
  • New Features:
    • Handoffs: Transfer control between agents (e.g., triage to specialized agents).
    • Guardrails: Input/output validation to prevent misuse.
    • Tracing & Observability: Debug workflows via execution traces.
• Enterprise Use Cases:
  • Hebbia: Uses web search for financial and legal research, improving market intelligence precision.
  • Navan: Leverages file search for travel policy queries, saving time for users.
  • Luminai: Automated legacy system workflows for a community service org using CUA, outperforming traditional RPA.
  • Coinbase & Box: Built crypto wallet agents and secure data analysis tools with the SDK.
  • Stripe: Developed an invoicing agent using the Agents SDK, streamlining payment workflows.
• Safety & Privacy:
  • CUA includes confirmation prompts for sensitive tasks, environment isolation, and policy violation detection.
  • OpenAI does not train models on business data stored on its platform, even when integrated.
• Assistants API Retirement: OpenAI plans to replace its Assistants API with the Responses API by mid-2026, incorporating developer feedback to improve functionality.
• Developer Resources: Both tools are available via OpenAI’s platform, expanding on the existing Chat Completions API.

Why This Matters:

OpenAI’s tools now offer quantifiable performance improvements and enterprise-grade safety, enabling industries to automate complex workflows (e.g., legacy system integration, real-time market analysis) with greater reliability. Benchmarks highlight progress toward human-level task execution, while pricing models make scalable AI agent deployment feasible. The launch aligns with 2025’s designation as the “year of AI agents,” further amplified by momentum from firms like China’s Manus, underscoring global demand for practical, customizable AI solutions.