OpenAI officially expanded its latest model family with the release of GPT-5.4 mini and GPT-5.4 nano. These “compact” models are designed to bring the advanced reasoning, coding, and multimodal capabilities of the flagship GPT-5.4 (released earlier this month) to high-volume, latency-sensitive applications.
GPT-5.4 Mini: The New “Mid-Tier” Powerhouse
OpenAI claims that GPT-5.4 mini is significantly more capable than its predecessor, running 2x faster than GPT-5 mini while nearing the performance of the full GPT-5.4 on several professional benchmarks.
- Coding Excellence: Scores 54.38% on SWE-bench Pro, a massive leap from the 45.69% achieved by the previous mini.
- Agentic Tool Use: Optimized for “subagent” workflows where a larger model (like GPT-5.4 Pro) plans a task and the mini model executes the technical sub-steps.
- Multimodal “Computer Use”: The model can rapidly interpret dense user interface screenshots to assist in autonomous computer navigation tasks.
- Context Window: Features a massive 400,000-token context window in the API.
GPT-5.4 Nano: High-Speed Intelligence at Scale
The GPT-5.4 nano is now OpenAI’s smallest and most affordable model. While designed for lightweight tasks like classification and data extraction, it has surprised early testers with its surprisingly high reasoning scores.
“GPT-5.4 nano is our most efficient model yet. It can describe roughly 76,000 photos for just $52, making it the ideal choice for massive-scale visual processing,” noted developer Simon Willison in a post-launch review.
Competitive Pricing & Availability
OpenAI has positioned these models to compete directly with Google’s Gemini 3.1 Flash-Lite and Anthropic’s Claude 4.5 Haiku.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Access |
| GPT-5.4 | $2.50 | $15.00 | API, Plus, Team, Enterprise |
| GPT-5.4 Mini | $0.75 | $4.50 | API, Codex, ChatGPT (Free/Go) |
| GPT-5.4 Nano | $0.20 | $1.25 | API Only |
ChatGPT Integration: * Free and “Go” Tier Users: GPT-5.4 mini is available via the “Thinking” option in the “+” menu.
- Paid Users: It serves as the primary rate-limit fallback for GPT-5.4 Thinking sessions.
The Codex “30% Quota” Rule
For developers using OpenAI Codex, GPT-5.4 mini has a unique advantage: it consumes only 30% of the standard GPT-5.4 quota. This allows engineers to handle codebase exploration and debugging loops at approximately one-third the cost of the flagship model.
Introducing GPT-5.4 mini and nano – OpenAI Official
This official announcement provides the full technical breakdown, benchmark comparisons (including OSWorld-Verified), and safety documentation for the new 5.4 small-model series.
