The consumer angle on GPT-5.4 Mini is already covered. This is the developer story.
OpenAI released GPT-5.4 mini and nano on March 18, positioning both as smaller, faster versions of GPT-5.4 built for coding, tool use, multimodal reasoning, and high-volume API workloads at lower cost than the full model. GPT-5.4 Nano is the smallest and fastest in the lineup, designed for low-latency, low-cost usage at high throughput, according to Microsoft’s Azure Tech Community documentation.
What does that mean in practice? Nano targets the tasks where speed matters more than depth: classification pipelines, content extraction, ranking signals, and lightweight tool-use chains in agentic workflows. Mini sits one tier up, better suited for multimodal reasoning and more complex coding tasks where a bit more latency is acceptable.
On benchmarks: according to data reported by DataCamp, GPT-5.4 mini scores 54.4% and nano scores 52.4% on SWE-Bench Pro, based on OpenAI’s evaluation data. Independent evaluation by Epoch AI or equivalent is pending. Treat these numbers as directional, not definitive.
Pricing is lower than the full GPT-5.4 model for both. No specific per-token figures are confirmed in available sources.
Developers currently using GPT-5 mini should run their own latency and cost benchmarks against both new models before migrating. The performance gap may be smaller than expected, but “vendor evaluation data” and “production performance on your workload” aren’t the same thing.