GPT-5.4 Mini and Nano: Fast and Cost-Effective AI Models

Introduction of GPT‑5.4 mini and nano Models

OpenAI has recently unveiled its latest artificial intelligence models, GPT‑5.4 mini and nano. These new versions are designed to deliver exceptional performance, particularly in the areas of programming and sub-agents. By integrating advancements from GPT‑5.4, these models stand out for their speed and efficiency, thus meeting the demands of large workloads.

Features of GPT‑5.4 mini

The GPT‑5.4 mini model represents a significant advancement over its predecessor, GPT‑5 mini. It excels in areas such as coding, reasoning, multimodal understanding, and tool usage. This model is not only more powerful but also more than twice as fast. It manages to compete with the larger model, GPT‑5.4, in several evaluations, including SWE-Bench Pro and OSWorld-Verified.

Focus on GPT‑5.4 nano

As for GPT‑5.4 nano, it is the most compact and cost-effective version of the GPT‑5.4 series. This model is ideal for tasks where speed and cost are critical factors. It represents a notable improvement over GPT‑5 nano and is recommended for applications such as classification, data extraction, ranking, as well as for programming sub-agents that handle simpler support tasks.

Performance and Applications

These models have been designed to excel in environments where latency is a determining factor for user experience. They are particularly suited for applications such as responsive coding assistants, sub-agents quickly executing support tasks, and systems capable of capturing and interpreting screenshots. In these contexts, model size does not always equate to better performance; rather, the ability to respond quickly and use tools reliably is paramount.

Performance Evaluation Results

The performance of the various models has been measured across several evaluations:

• SWE-Bench Pro (Public)

  • GPT-5.4: 57.7%
  • GPT-5.4 mini: 54.4%
  • GPT-5.4 nano: 52.4%
  • GPT-5 mini: 45.7%

• Terminal-Bench 2.0

  • GPT-5.4: 75.1%
  • GPT-5.4 mini: 60.0%
  • GPT-5.4 nano: 46.3%
  • GPT-5 mini: 38.2%

• OSWorld-Verified

  • GPT-5.4: 75.0%
  • GPT-5.4 mini: 72.1%
  • GPT-5.4 nano: 39.0%
  • GPT-5 mini: 42.0%

User Feedback

Initial feedback from users who have integrated GPT‑5.4 mini and nano into their workflows is positive. Aabhas Sharma, CTO at Hebbia, stated that GPT-5.4 mini offers solid performance for a model in its category. In evaluations, it matched or surpassed competing models on several generation and citation recall tasks while being more economical. It also demonstrated increased reliability in source attribution compared to the larger model, GPT-5.4.

Efficiency in Programming Processes

The GPT‑5.4 mini and nano models are particularly effective in programming processes that require rapid iterations. They efficiently handle targeted modifications, code navigation, front-end generation, and debugging loops with low latency. This makes them an ideal option for coding tasks that require speed and reduced costs.

Availability and Costs

GPT‑5.4 mini is now available via API, Codex, and ChatGPT. In the API, it supports text and image inputs, tool usage, function calling, web searching, file searching, computer usage, and skills. It has a context window of 400k and costs $0.75 per 1M tokens for input and $4.50 per 1M tokens for output.

In Codex, GPT‑5.4 mini is available in the Codex app, CLI, IDE extension, and on the web, using only 30% of the quota of GPT‑5.4, allowing developers to quickly handle simpler coding tasks for about a third of the cost.

In ChatGPT, GPT‑5.4 mini is available for Free and Go users through the "Thinking" feature in the + menu. For other users, it is offered as a fallback solution in case of throttling for GPT‑5.4.

GPT‑5.4 nano is only accessible via the API, with a cost of $0.20 per million tokens for input and $1.25 per million tokens for output.