OpenAI has released GPT-5.4 mini and nano, two new smaller models designed for high-volume workloads where speed and cost are critical.
Key Features and Improvements
- GPT-5.4 mini:
- Significantly improves upon GPT-5 mini in coding, reasoning, multimodal understanding, and tool use.
- Runs more than 2x faster than GPT-5 mini.
- Approaches the performance of the larger GPT-5.4 model on evaluations like SWE-Bench Pro and OSWorld-Verified.
- GPT-5.4 nano:
- The smallest and cheapest version of GPT-5.4.
- Suitable for tasks where speed and cost are paramount.
- Recommended for classification, data extraction, ranking, and coding subagents handling simpler tasks.
Use Cases
These models are designed for applications where latency directly impacts the user experience, such as:
- Coding assistants
- Subagents
- Computer-using systems interpreting screenshots
- Real-time multimodal applications
The article emphasizes that the best model isn’t always the largest, but the one that responds quickly, uses tools reliably, and performs well on complex tasks.
Performance Benchmarks
The article includes performance benchmarks comparing GPT-5.4, GPT-5.4 mini, GPT-5.4 nano, and GPT-5 mini on various tasks, including:
- SWE-Bench Pro
- Terminal-Bench 2.0
- Toolathlon
- GPQA Diamond
- OSWorld-Verified
These benchmarks demonstrate the improved performance of GPT-5.4 mini and nano compared to their predecessors.
Customer Feedback
Early customer testing indicates that GPT-5.4 mini and nano are particularly effective in coding workflows that benefit from fast iteration, such as targeted edits, codebase navigation, front-end generation, and debugging.
System Composition
The article suggests a system architecture where larger models like GPT-5.4 handle planning and coordination, while smaller models like GPT-5.4 mini execute subtasks in parallel.
Multimodal Capabilities
GPT-5.4 mini is also strong in multimodal tasks, especially those related to computer use, such as interpreting screenshots.
Availability and Pricing
- GPT-5.4 mini: Available in the API, Codex, and ChatGPT.
- Supports text and image inputs, tool use, function calling, web search, file search, computer use, and skills.
- 400k context window.
- Pricing: $0.75 per 1M input tokens and $4.50 per 1M output tokens.
- GPT-5.4 nano: Available only in the API.
- Pricing: $0.20 per 1M input tokens and $1.25 per 1M output tokens.
Performance Data
The article provides detailed performance data across various categories:
- Coding
- Tool-calling
- Intelligence
- MM / Vision / CUA
- Long context
These tables compare the performance of GPT-5.4, GPT-5.4 mini, GPT-5.4 nano, and GPT-5 mini on specific benchmarks within each category.