OpenAI Introduces GPT-54 Mini Nano as Faster Models Optimised for Coding and AI Agents

OpenAI introduced two new artificial intelligence (AI) models in the GPT-5, a model that was previously unveiled by Open AI. Tuesday, 4 family with 4. GPT-5 – known as . 5 PT-5 and 4 mini. Both small AI models are faster than the larger models in family, 4 nano and target low-latency workloads (i.e. The key strengths of these models are coding proficiency, computer use, multimodal understanding and subagent handling. Such models are also cost-efficient for developers, as input and output tokens will be cheaper.

OpenAI Introduces GPT-5.4 Mini and GPT-5.4 Nano

San Francisco-based AI giant announced the launch of two new models in a blog post on Wednesday that the two model models will soon be released. GPT5 – . Now the application programming interface (API), Codex and ChatGPT have access to 4 mini. The 400,000 tokens context window allows text and image inputs, tool use, function calling, web and file search (computer use) and skills from the model in the API. It costs $0 for . A total of 75 per million input tokens and $4 are also issued for . 50 per million output tokens.

The API uses GPT5 as an example of . 4 mini Text and image inputs, tool use, function calling, web search, file search (computer usage) skills skills. It has a 400k context window and costs $0. 75 (roughly Rs. For example, 68) per 1M input token and $4. Paraphrasing 50 (roughly Rs. 1M output tokens 416) per .

Notably GPT-5, . During the free and Go tiers (which use The Thinking option) 4 mini is available to the Free, while other levels will consider it as a fallback model when they reach the rate limit for GPT-5. 4-thinking – 4 Thinking. Visiting GPT-5 . This is now a sole API service, with pricing at $0 for 4 nano. One-million dollars, 20 per million input and $1. Per million output tokens are 25 per million for .

Similarly to both models, on capabilities, they are optimised for coding-related tasks as long as the model is deployed in fast, iterative environments. In OpenAI, the models “handle targeted edits,” codebase navigation, front-end generation and debugging loops with low latency”, it says. Moreover, the 5 is said to have been an . Most areas at similar latencies have 4 mini that are better than GPT-5-mini.

A second model’s strength is subagent handling, another of the models unique strengths. Despite the larger AI models in the family, which are suitable for more complex agentic tasks such as planning, coordination and final judgment, the mini variant can handle subagents that care about narrower subtasks in parallel with smaller ones.

As for the smaller models of OpenAI, these are able to write systems where one single model is not taking advantage of all subtasks in an agentic workflow. Besides this, the company says that the mini variant also performs well on multimodal tasks such as computer use. Interestingly, on the OSWorld-Verified benchmark, the mini variant approaches GPT-5. 4 .