GPT-5.4 mini and nano deliver near-flagship performance at lower cost

OpenAI has released GPT-5.4 mini and GPT-5.4 nano, smaller models designed for fast, efficient, high-volume AI workloads such as responsive coding assistants, subagents, screenshot interpretation, and multimodal applications. The company says GPT-5.4 mini runs more than twice as fast as GPT-5 mini and is built for settings where latency directly shapes the product experience.

Benchmarks show notable gains for the mini model: 54.38% on SWE-bench Pro versus 45.69% for GPT-5 mini, 60.00% on Terminal-Bench 2.0 versus 38.20%, and 88.01% on GPQA Diamond, approaching GPT-5.4's 93.00%. OSWorld-Verified results put GPT-5.4 mini at 72.13% versus GPT-5 mini's 42%.

GPT-5.4 nano scores 52.39% on SWE-bench Pro and 46.30% on Terminal-Bench 2.0, trailing the mini but still outperforming GPT-5 mini on several tasks. Early customer testing highlights practical benefits.

openai, gpt-5.4 mini, gpt-5.4 nano, gpt-5 mini, swe-bench pro, terminal-bench 2.0, gpqa diamond, osworld-verified, latency, coding assistants