OpenAI launches GPT-5.4 Mini and Nano models following GPT-5.4 debut
Sam Altman's OpenAI announced the release of GPT-5.4 mini and GPT-5.4 nano on Wednesday, describing them as its most advanced compact AI models to date. The launch comes shortly after the company rolled out its GPT-5.4 AI model earlier this week. OpenAI has positioned GPT-5.4 as its leading model for professional workflows across its chatbot ChatGPT and coding platform Codex.
The flurry of model releases comes as the generative AI space becomes more competitive. The company had been fielding tough competition from Google's Gemini models last year. The beginning of this year propelled Anthropic's models to the top of the charts, after the Claude maker stood up to the US government.
The new model GPT-5.4 mini offers improvements over the GPT-5 mini in various areas, including coding, reasoning, multimodal tasks, and tool use, the company said in a statement. The model runs more than twice as fast while approaching GPT-5.4 performance levels on several benchmarks. GPT-5.4 nano, the smallest variant, targets simpler tasks such as classification, data extraction, ranking, and basic coding support.
OpenAI has touted the new models for applications where response speed is key, such as coding assistants, subagents for supporting tasks, screenshot interpretation, and real-time image analysis.
In internal benchmarks, GPT-5.4 mini reportedly outperforms GPT-5 mini at similar speeds, providing what the company calls a strong performance-to-latency balance.
Within Codex, the new models support hybrid setups. GPT-5.4 can manage planning and oversight, while GPT-5.4 mini handles subtasks such as codebase searches or file reviews, the statement said. OpenAI notes this uses 30% of GPT-5.4 quotas at about one-third the cost.
The flurry of model releases comes as the generative AI space becomes more competitive. The company had been fielding tough competition from Google's Gemini models last year. The beginning of this year propelled Anthropic's models to the top of the charts, after the Claude maker stood up to the US government.
The new model GPT-5.4 mini offers improvements over the GPT-5 mini in various areas, including coding, reasoning, multimodal tasks, and tool use, the company said in a statement. The model runs more than twice as fast while approaching GPT-5.4 performance levels on several benchmarks. GPT-5.4 nano, the smallest variant, targets simpler tasks such as classification, data extraction, ranking, and basic coding support.
OpenAI has touted the new models for applications where response speed is key, such as coding assistants, subagents for supporting tasks, screenshot interpretation, and real-time image analysis.
In internal benchmarks, GPT-5.4 mini reportedly outperforms GPT-5 mini at similar speeds, providing what the company calls a strong performance-to-latency balance.
Within Codex, the new models support hybrid setups. GPT-5.4 can manage planning and oversight, while GPT-5.4 mini handles subtasks such as codebase searches or file reviews, the statement said. OpenAI notes this uses 30% of GPT-5.4 quotas at about one-third the cost.
Next Story