OpenAI launches GPT-5.4 mini and GPT-5.4 nano, with performance approaching that of the full-fledged version.

OpenAI recently released two lightweight models: GPT-5.4 mini and GPT-5.4 nano. GPT-5.4 mini achieved a 54.4% score in the SWE-Bench Pro coding benchmark, 3.3 percentage points lower than the full-fledged GPT-5.4, while running twice as fast as its predecessor. It also supports ultra-long contexts of up to 400,000 tokens, with an input cost of only $0.75 per million tokens and an output cost one-third that of the full-fledged model. GPT-5.4 nano is positioned for ultra-lightweight tasks, priced at only 1/12th of the full-fledged model.