OpenAI Launches GPT-5.4 Mini and Nano

🔑 Enhanced Key Takeaways

• •GPT-5.4 mini features a 400k context window in the API, enabling handling of extensive inputs for complex tasks.[2]
• •GPT-5.4 nano is priced at $0.20 input / $1.25 output per 1M tokens via API, making it the cheapest GPT-5.4 variant.[2]
• •GPT-5.4 mini provides ~3.3x more usage on Codex tasks compared to full GPT-5.4, ideal for intensive development workflows.[2]
• •Both models support vision inputs up to 1600x1600 pixels billed as 2500 patches, despite documentation discrepancies.[2]

🛠️ Technical Deep Dive

• •GPT-5.4 mini runs more than 2x faster than GPT-5 mini while improving in coding, reasoning, multimodal understanding, and tool use; approaches full GPT-5.4 on SWE-Bench Pro and OSWorld-Verified benchmarks.[1][2]
• •400k token context window available for GPT-5.4 mini in the API.[2]
• •GPT-5.4 nano optimized for classification, data extraction, ranking, and simple coding subagents.[1][2]
• •API pricing for GPT-5.4 nano: $0.20 per 1M input tokens / $1.25 per 1M output tokens.[2]
• •Vision capabilities support up to 1600x1600 pixel images (2500 patches/tokens), with noted discrepancies in resizing and billing for Chat Completions.[2]

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline