OpenAI unveiled GPT‑5.4 mini and nano—compact variants of the main model designed for high-load tasks.

OpenAI releases compact versions of GPT‑5.4

The company announced the launch of two new small AI models – GPT‑5.4 mini and GPT‑5.4 nano, which combine the key capabilities of full‑size GPT‑5.4 but run faster and require fewer resources.

What’s new?
ModelKey FeaturesWhere It HelpsGPT‑5.4 mini• Improved code generator
• More logical reasoning
• Third‑party tool management
• Speed more than twice that of GPT‑5 mini
• Results at full‑size GPT‑5.4 level on SWE‑Bench Pro and OSWorld‑Verified tests• Fast AI assistants for code
• Sub‑agents performing auxiliary tasks
• Multimodal applications (screen‑shot recognition, UI control)
• Chatbots (when GPT‑5.4 Thinking limits are exceeded in ChatGPT, it automatically switches to mini)GPT‑5.4 nano• The most compact and fastest model in the GPT‑5.4 family
• Significantly outperforms GPT‑5 nano• Data extraction, classification, and ranking
• Sub‑agents for basic tasks

How It Works in Practical Scenarios
- Codex: The senior GPT‑5.4 model coordinates parallel sub‑agents based on GPT‑5.4 mini. This allows, for example, searching code in a repository, analyzing large files, and working with documentation – all without using the heaviest model.

- Development: Teams use the “large” model to make decisions, while small models quickly implement those decisions on large data volumes. GPT‑5.4 mini also efficiently manages the PC user interface.

Access and Pricing
ModelWhere AvailableToken LimitPrice (input)Price (output)GPT‑5.4 miniCodex app, CLI, IDE extension, web interface, API400,000 tokens$0.75/million$4.50/millionQuota consumption: 30 % of GPT‑5.4GPT‑5.4 nanoOnly API—$0.20/million$1.25/million

Thus, OpenAI has expanded its AI solutions portfolio by providing fast and cost‑effective options for tasks where speed and efficiency are more important than full model power.