GPT‑5.4 mini & nano
OpenAI released GPT‑5.4 mini and GPT‑5.4 nano—mini brings improved reasoning and multimodal smarts to ChatGPT Free, Go and GitHub Copilot, while nano is API‑only and billed ultra‑cheap at roughly $0.20 per million input tokens. The move pushes stronger small‑model baselines for mobile and backend integrations. (9to5mac.com) (github.blog)
GPT‑5.4 mini is listed at $0.75 per 1M input tokens and $4.50 per 1M output tokens in the OpenAI model docs. (developers.openai.com) Both GPT‑5.4 mini and the smaller variant expose a 400k‑token context window in the API, enabling single‑request access to very large codebases or document collections. (openai.com) OpenAI’s public benchmark table shows SWE‑Bench Pro scores of 57.7 for GPT‑5.4, 54.4 for GPT‑5.4 mini, and 52.4 for GPT‑5.4 nano, and Terminal‑Bench 2.0 scores of 75.1, 60.0, and 46.3 respectively. (openai.com) GitHub’s changelog notes GPT‑5.4 mini is rolling out in Copilot across VS Code, JetBrains, Xcode, the web, mobile and CLI, appears with a 0.33x premium request multiplier on launch, and requires admins to enable the GPT‑5.4 mini policy for orgs or use BYOK to bind an API key. (github.blog) OpenAI positions mini for low‑latency, tool‑heavy coding and “computer‑use” subagent work with support for function calling, web and file search, and skills, while recommending the smaller model for high‑throughput classification, extraction, and ranking tasks. (openai.com) Operational cost levers include a 10% uplift for regional/data‑residency endpoints and a Batch API option that OpenAI says can save roughly 50% on inputs and outputs for asynchronous, high‑volume jobs. (developers.openai.com)
