OpenAI drops GPT-5.4 mini, nano for agents

OpenAI is aggressively optimizing for the "subagent era" by decoupling high-level reasoning from high-volume execution tasks required for autonomous workflows. GPT-5.4 mini delivers a 2x speed boost over its predecessor while maintaining flagship-level performance on computer-use benchmarks like OSWorld, while the "nano" variant introduces extreme cost efficiency at $0.20 per million tokens. A 400k context window in these smaller models facilitates deep codebase navigation and RAG-heavy agent tasks without the latency of larger flagship models. The new tiered UI signals a shift toward intent-based model selection, abstracting technical names behind user-centric labels like "Thinking" and "Instant."