OpenAI has announced the release of the GPT-5.4 Mini and Nano models, expanding its free-to-use offerings. These new models follow the launch of the more powerful GPT-5.4, providing smaller and more cost-effective alternatives for users prioritizing speed and affordability.
The GPT-5.4 Mini boasts enhanced performance in tasks such as coding, reasoning, multimodal understanding, and tool use. It runs over twice as fast as its predecessor and demonstrates competitive performance compared to the GPT-5.4 on specific benchmarks like SWE-Bench Pro and OSWorld-Verified.
The GPT-5.4 Nano is positioned as the most compact and economical version of the GPT-5.4 series, catering to scenarios prioritizing speed and cost efficiency. This model is well-suited for tasks like classification, data extraction, ranking, and coding subagents.
OpenAI emphasizes that these models are designed for workloads where latency significantly influences user experience, such as responsive coding assistants, subagents for swift task completion, systems interpreting screenshots, and real-time multimodal reasoning over images.
The introduction of the GPT-5.4 Mini as a subagent allows developers to offload specific tasks to this model while the primary GPT-5.4 focuses on planning, coordination, and final decisions. This setup enables efficient handling of narrower, parallel subtasks like codebase searches or document reviews.
OpenAI highlights the GPT-5.4 Mini’s proficiency in complex computer-use tasks, including rapid interpretation of dense user interface screenshots and real-time image reasoning. Moreover, it narrows the performance gap with the flagship GPT-5.4 on visual benchmarks like OSWorld-Verified.
Source: mint – technology