Two New Lightweight Models from OpenAI
OpenAI presented its two new compact models, GPT-5.4 mini and GPT-5.4 nano, on March 17, 2026. According to the company, these are the most capable small models they have released to date, optimized for coding, tool use, multimodal reasoning, and high-volume API workflows.
Both models are leaner and faster variants of the already launched GPT-5.4, and are specifically designed for sub-agent tasks and large-scale developer use.
GPT-5.4 mini is more than twice as fast as GPT-5 mini and approaches the performance of the full model on key benchmarks
GPT-5.4 mini: Near Full-Model Performance at Lower Cost
The larger of the two newcomers, GPT-5.4 mini, scores 54.4 percent on SWE-Bench Pro – a recognized benchmark for code agents. In comparison, the full GPT-5.4 model achieves 57.7 percent, while the older GPT-5 mini was at 45.7 percent. On OSWorld-Verified, which tests agents' ability to operate computers, mini lands at 72.1 percent against the full model's 75.0 percent, according to OpenAI.
The model supports text and image input, function calling, web search, file search, and computer control via the API. The context window is 400,000 tokens, with a maximum output of 128,000 tokens. The knowledge cutoff is set to August 31, 2025.
In the Codex platform, GPT-5.4 mini uses only 30 percent of the GPT-5.4 quota, which practically provides access to code assistance at approximately one-third of the price.
GPT-5.4 nano: Ultra-Low Price for High Volume
The smallest model, GPT-5.4 nano, is exclusively available via API and is designed for tasks where speed and cost are paramount – such as classification, data extraction, ranking, and simple sub-agent functions without the need for multi-step reasoning.
Despite its compact size, nano scores 52.4 percent on SWE-bench Pro, which actually surpasses GPT-5 mini's 45.7 percent. It is worth noting that these benchmark figures originate from OpenAI itself, and independent verification is not available at the time of publication.
Price Overview and Availability
GPT-5.4 mini is available via the OpenAI API, Codex, and ChatGPT. In ChatGPT, free and Go users can utilize the model through the “Thinking” function in the plus menu, and it also serves as a fallback model for rate limiting on GPT-5.4 Thinking for other subscribers. GPT-5.4 nano is currently only available via API.
What This Means for Developers
With these two models, OpenAI is trying to meet a growing market need: developers who want modern model capabilities – especially in coding and agent workflows – without paying for full GPT-5.4 capacity on all requests.
Combined with the large context window of 400,000 tokens and support for multimodal input, GPT-5.4 mini appears as an attractive alternative for production systems where both performance and operating costs matter. Nano fills another need: classification and simple extraction tasks on an industrial scale, where the price per token is crucial.
The OpenAI platform runs on Microsoft Azure, and all API access requires an API key created through the OpenAI dashboard.