GPT-5.4 mini

OpenAI released GPT-5.4 mini on March 17, 2026 as a fast, efficient small model optimized for coding, reasoning, multimodal understanding, and tool use in high-volume workloads. According to the announcement, it improves over GPT-5 mini while running more than 2x faster, and on several evaluations it approaches the larger GPT-5.4 model's performance. Reported benchmark scores include 54.4% on SWE-Bench Pro, 60.0% on Terminal-Bench 2.0, 42.9% on Toolathlon, 88.0% on GPQA Diamond, and 72.1% on OSWorld-Verified (all xhigh reasoning effort). Use cases highlighted include responsive coding assistants, supporting subagents, and computer-using systems that interpret screenshots in real time. Hebbia CTO Aabhas Sharma stated in the announcement that GPT-5.4 mini matched or exceeded competitive models on output tasks and citation recall at a much lower cost, and achieved stronger source attribution than the larger GPT-5.4. The model is positioned as the workhorse tier of the GPT-5.4 family, complementing the smallest, cheapest GPT-5.4 nano variant for tasks where speed and cost dominate. OpenAI recommends it for high-volume production paths where latency, reliable tool use, and complex professional performance must coexist without using the largest model.

GPT-5.4 mini launched on March 17, 2026 and is available in ChatGPT, Codex, and via the OpenAI API, supporting text and image inputs, tool use, function calling, web search, file search, computer use, and skills. The model features a 400k context window and is priced at $0.75 per 1M input tokens and $4.50 per 1M output tokens, up from GPT-5 mini's $0.25/$2. In Codex (app, CLI, IDE, and web), GPT-5.4 mini consumes only 30% of the GPT-5.4 quota, making it suitable for less reasoning-intensive work. ChatGPT Free and Go users can access GPT-5.4 mini via the Thinking feature in the + menu, while premium plan users get it as a fallback for GPT-5.4 Thinking. The article notes GPT-5.4 mini nearly matches GPT-5.4 on SWE-Bench Pro and OSWorld-Verified, and outperforms GPT-5 mini across coding, reasoning, multimodal understanding, and tool use. GPT-5.4 nano is available via API only, not in ChatGPT or Codex, filling the cheapest tier for classification, extraction, ranking, and coding subagents.

The apxml.com catalog page for GPT-5.4 mini lists it as a closed-source, proprietary OpenAI model released on March 16, 2026, with an undisclosed parameter count and architecture, multimodal input, a 400K context length, and API pricing of $0.75 per 1M input tokens and $4.50 per 1M output tokens. The page notes that GP According to the same page, GPT-5.4 mini supports text and image inputs plus tool use, function calling, web search, file search, computer use, and skills, and is available via the API as gpt-5.4-mini. The page also catalogues published benchmark ranks across GPQA, LiveBench categories, Text Arena, WebDev Arena, and Ar

Independent EdTech Innovation Hub coverage dated 19 March 2026 reports that OpenAI released GPT-5.4 mini and GPT-5.4 nano, two smaller models designed for high-volume workloads balancing performance, cost, and latency, announced via a LinkedIn post from OpenAI for Business. The article quotes OpenAI saying: "Today we'r According to the same article, GPT-5.4 mini improves on GPT-5 mini across coding, reasoning, multimodal understanding, and tool use while running more than twice as fast, and approaches the performance of the larger GPT-5.4 model on SWE-Bench Pro and OSWorld-Verified. GPT-5.4 nano is positioned as the smallest, lowest-