GPT‑5.4 Mini & Nano
OpenAI rolled out GPT‑5.4 ‘Mini’ and ‘Nano’ models designed for faster, lower‑cost coding and doc tasks—Mini promises near‑flagship performance while Nano targets large‑scale, cheap workloads. Teams can use these for automated reporting, code‑review summaries, and multimodal dashboard interpretation, though accuracy caveats remain. (gbhackers.com)
OpenAI published GPT‑5.4 mini and nano on March 17, 2026 and made both models available in ChatGPT, the API, and Codex. (openai.com) OpenAI’s benchmark table shows SWE‑Bench Pro scores of 57.7% for GPT‑5.4 (xhigh), 54.4% for GPT‑5.4 mini, 52.4% for GPT‑5.4 nano, and 45.7% for GPT‑5 mini. (openai.com) Terminal‑Bench 2.0 in the same table reports 75.1% for GPT‑5.4 versus 60.0% for GPT‑5.4 mini and 46.3% for GPT‑5.4 nano, illustrating where mini/nano trade some peak capability for speed. (openai.com) OpenAI and the developer forum state GPT‑5.4 mini runs more than 2× faster than GPT‑5 mini and the models expose a 400k token API context window for long‑context workloads. (openai.com) API list pricing published in the community post lists GPT‑5.4 mini at $0.75 input / $4.50 output per 1M tokens and GPT‑5.4 nano at $0.20 input / $1.25 output per 1M tokens, with Codex consumption noted against a 30% GPT‑5.4 quota. (community.openai.com) Microsoft’s Foundry team confirmed a same‑day rollout in the Foundry model catalog and positioned the models for planner+executor agent patterns, low‑latency multimodal tooling, and high‑throughput classification/extraction workloads. (techcommunity.microsoft.com) OpenAI published a list of early testers including CodeRabbit, Mercor, GitHub, Rox, Notion, Whoop, and Perplexity, and quoted Aabhas Sharma (CTO at Hebbia) saying GPT‑5.4 mini “matched or exceeded competitive models on several output tasks and citation recall at a much lower cost.” (openai.com) OpenAI’s documentation explicitly warns that simulated latency estimates do not capture all production variables and that “real‑world latency may vary substantially” depending on tool call durations, sampled tokens, and other deployment factors. (openai.com)
