GPT-4o, GPT-5, o-series reasoning models, the Responses API, Realtime API, Codex CLI, and the Swarm agents library. The pace-setter on frontier capability and the default starting point for any new AI architecture decision.
GPT-4o, GPT-5, o-series — reasoning + multimodal
GPT-4o (omni — text/audio/vision), GPT-5 frontier reasoning, o1/o3 reasoning-optimized variants, GPT-4o mini for cost-tier serving. Model selection is per-task: o-series for complex multi-step reasoning, GPT-4o for multimodal + general tasks, mini variants for throughput-sensitive workloads. Function calling and JSON mode are first-class.
Chat Completions, Responses API, Assistants, Realtime
The API surface: Chat Completions (legacy + structured outputs), the newer Responses API (server-side state + tool use), Assistants API (managed threading + retrieval), Realtime API (voice + low-latency multimodal), Batch API (50% discount, async). Streaming, function calling, structured outputs, vision, and audio all first-class.
OpenAI's coding-optimized agent surface
OpenAI's code-specialized model lineage from the original Codex through the current code-tuned GPT variants and the standalone Codex CLI. Pairs naturally with the LLM Council pattern as the adversarial code-review counterpart to Claude Code's orchestrator role. The interesting positioning: terminal-resident agent vs IDE-resident (Cursor / Copilot).
Lightweight ergonomic agent orchestration library
OpenAI's minimal agents framework — released as educational/research code, since superseded by Agents SDK but the design principles persist. Handoffs between agents, function calling as the routing primitive, no heavyweight state machine. Compares directly to CrewAI (more opinionated) and LangGraph (more control). The right call when you want agents without ceremony.