GPT-5.4 mini - OpenAI's most powerful small model

What is GPT-5.4 mini

GPT-5.4 mini is OpenAI's most powerful small model, designed for high-throughput, low-latency scenarios. The model is more than 2x faster than the previous generation GPT-5 mini, with performance approaching the full version in tasks such as coding, reasoning, multimodal understanding, and computer useGPT-5.4, at only 1/3 of its price. GPT-5.4 mini supports a 400K context window and is suitable for scenarios such as coding assistants, sub-agents, and real-time image parsing. It is now available on the API, Codex, and ChatGPT, making it an ideal choice for building multi-agent systems.

GPT-5.4 mini's main features

• Coding and development: Supports targeted code editing, codebase navigation, front-end generation, and debugging loops, achieving a 54.4% pass rate on the SWE-Bench Pro benchmark, approaching the level of the full GPT-5.4.
• Multimodal understanding: The model can quickly parse complex user interface screenshots and perform real-time visual reasoning, scoring 78% on the MMMUPro benchmark.
• Computer use: The model has strong capabilities in screen content parsing and interface element localization, reaching 72.1% on the OSWorld-Verified benchmark, only 3 percentage points lower than GPT-5.4.
• Tool calling: Supports function calling, web search, file search, and skill execution, scoring 57.7% on the MCP Atlas tool calling benchmark.
• Sub-agent tasks: Can act as a sub-agent to execute tasks such as searching codebases, reviewing large files, and processing auxiliary documents in parallel, consuming only 30% of the GPT-5.4 quota in Codex.
• Long context processing: Supports an ultra-large context window of 400K tokens, capable of handling long document analysis and complex multi-turn conversations.

GPT-5.4 mini's key information and usage requirements

• Positioning: OpenAI's most powerful small model, designed for high-throughput, low-latency scenarios
• Speed: More than 2x faster than GPT-5 mini
• Performance: Close to GPT-5.4 in tasks such as coding, reasoning, multimodal, and computer use
• Context：400K tokens
• Pricing: Input $0.75 per million tokens, output $4.50 per million tokens (about 1/3 of GPT-5.4)
• Access channels：API、Codex、ChatGPT

GPT-5.4 mini's core advantages

• Both speed and performance: More than 2x faster than GPT-5 mini, while performance in tasks such as coding, reasoning, and multimodal is close to the full GPT-5.4.
• Extreme cost-effectiveness: Output price is only 1/3 of GPT-5.4, and it consumes only 30% of the quota in Codex, greatly reducing development costs.
• Strong multimodal capabilities: Quickly parses complex UI screenshots, reaching 72.1% on the OSWorld-Verified computer use benchmark, close to GPT-5.4's 75%.
• Ultra-large context: Supports a 400K tokens context window, capable of handling long documents and complex multi-turn conversations.
• Sub-agent optimization: Designed for parallel task execution, suitable as a sub-agent for auxiliary tasks such as search and review, building a combined architecture of large-model decision-making + small-model execution.
• Full platform coverage: Now available on the API, Codex, and ChatGPT, supporting complete capabilities such as text/image input, tool calling, function calling, web search, file search, and computer use.

How to use GPT-5.4 mini

• API usage: Called via the OpenAI API, it supports text and image input, tool use, function calling, web search, file search, computer use, and skill invocation, with a context window of up to 400K tokens.
• Codex usage: The model is fully available in Codex app, CLI, IDE extensions, and the web. It consumes 30% of the GPT-5.4 quota by default and can be configured as the default model for subagents, used for low-reasoning-effort tasks such as searching codebases, reviewing large files, and processing auxiliary documents.
• ChatGPT usage: Free and Go users access it directly through the "Thinking" feature in the "+" menu. Other users automatically switch to it as a rate-limit fallback for GPT-5.4 Thinking.

Comparison of GPT-5.4 mini with similar competitors

Application scenarios of GPT-5.4 mini

• Content classification: Quickly label and classify massive amounts of text and images and perform sentiment analysis, suitable for social media content moderation, news topic categorization, and user comment filtering.
• Data extraction: Extract structured data in bulk from unstructured documents, web pages, and tables, suitable for resume parsing, invoice information extraction, and identification of key contract fields.
• Ranking and filtering: Score relevance and prioritize search results, recommended content, and candidate lists, suitable for e-commerce product recommendations, recruitment resume screening, and personalized information feeds.
• Lightweight subagents: Serve as subagents to perform edge tasks such as validation, formatting, and simple queries, working with GPT-5.4/mini to build low-cost multi-agent systems.