OpenAI has introduced GPT-5.4 Mini and GPT-5.4 Nano, its latest lightweight AI models designed to deliver high performance at lower cost and latency. The new models bring many of the capabilities of the flagship GPT-5.4 into smaller, more efficient versions optimized for high-volume workloads.
The launch reflects a growing shift in the AI industry toward smaller, faster models that can handle everyday tasks at scale without the heavy compute costs of frontier models.
Built for Speed, Scale, and Cost Efficiency
Both GPT-5.4 Mini and Nano are designed for high-throughput applications such as chatbots, coding assistants, and automated workflows.
According to OpenAI:
GPT-5.4 Mini delivers strong reasoning and coding performance while being significantly cheaper than the full GPT-5.4 model
GPT-5.4 Nano is the smallest and most cost-efficient variant, optimized for lightweight tasks like classification and data extraction
The models are particularly suited for “subagent” workflows, where multiple smaller AI agents handle parts of a larger task to reduce costs.
Strong Performance Despite Smaller Size
Despite their reduced size, the new models retain much of the intelligence of GPT-5.4.
On key benchmarks:
GPT-5.4 Mini scores 54.4% on SWE-Bench Pro, close to GPT-5.4’s 57.7%
GPT-5.4 Nano achieves 52.4%, outperforming earlier lightweight models
Both models show strong performance in reasoning, tool usage, and coding tasks
These results suggest that smaller models are rapidly closing the gap with larger frontier systems.
Optimized for Coding and Agent Workflows
OpenAI is positioning the new models as ideal for developer-centric workflows, especially in tools like Codex.
Capabilities include:
Code generation and debugging
Navigating large codebases
Performing targeted edits
Supporting AI agents and automation loops
GPT-5.4 Mini is also 2× faster than GPT-5 Mini, making it more suitable for real-time coding assistance and iterative development.
Availability Across ChatGPT, API, and Codex
OpenAI is rolling out the models across its ecosystem:
GPT-5.4 Mini
Available in ChatGPT (including free and Go tiers)
Available in API and Codex
GPT-5.4 Nano
Available via API only
Priced at $0.20 per 1M input tokens and $1.25 per 1M output tokens
Mini also acts as a fallback or lower-cost option when higher-tier models reach usage limits.
A Shift Toward Multi-Model AI Architectures
The release highlights OpenAI’s broader strategy of tiered AI systems, where different models handle different workloads:
Large models → complex reasoning and high-value tasks
Mini/Nano models → high-volume, routine operations
This architecture allows companies to optimize both cost and performance by assigning tasks to the most appropriate model.
Source: OpenAI
Join the conversation! Your thoughts help the community grow.