OpenAI Releases GPT-5.4 Mini and Nano: Faster, More Affordable AI

OpenAI has further expanded its artificial intelligence model lineup with the release of two new models, GPT-5.4 Mini and GPT-5.4 Nano, on Tuesday, 17 March 2026.

The arrival of these two models continues a series of AI model updates for ChatGPT, which has accelerated considerably in recent months.

On 5 March, OpenAI released GPT-5.4 Thinking, a model featuring deep reasoning capabilities. Two days prior, the company also launched GPT-5.3 Instant, a model focused on everyday conversation to ensure smoother interactions, although not always with greater accuracy.

Of the two new models, GPT-5.4 Mini stands out most prominently.

On its official website, the AI company founded by Sam Altman states that this model delivers performance approaching the full GPT-5.4 version, but with significantly higher speeds.

GPT-5.4 Mini is claimed to operate more than twice as fast as GPT-5 Mini.

In various tests, the model has demonstrated significant improvements over GPT-5 Mini, ranging from coding capabilities to reasoning and multimodal processing such as image handling.

On Terminal-Bench 2.0, the model achieved 60 per cent, far exceeding GPT-5 Mini’s 38.20 per cent. This test assesses AI capabilities in using command line or terminal functions commonly employed by developers.

Similar results emerged in OSWorld-Verified, where GPT-5.4 Mini achieved 72.13 per cent, a significant increase compared to GPT-5 Mini’s 42 per cent.

This test evaluates AI capabilities in performing computer-based tasks, such as understanding screen displays and executing specific actions.

Meanwhile, on GPQA Diamond, which tests high-level reasoning capabilities, particularly complex science questions, GPT-5.4 Mini scored 88.01 per cent, approaching the flagship model GPT-5.4’s 93 per cent.

Overall, these figures demonstrate that GPT-5.4 Mini has nearly achieved the success rate of the flagship GPT-5.4 model, but with faster execution.