OpenAI GPT-5.4 Mini Text + Vision

Overview

OpenAI built GPT-5.4 mini as a fast GPT‑5.4‑class model for high-volume work. You give it text instructions and, when needed, images to analyze. It can apply deeper or lighter reasoning based on your chosen effort level. It returns text, which makes it easy to drop into apps that need answers, summaries, or code.

What you can build

• A coding assistant that writes, edits, and explains code with consistent style rules.
• A “screenshot helper” that reads UI screenshots and tells users what to click next.
• A document intake flow that extracts key fields from forms, tables, and screenshots.
• A support agent that triages issues from error screenshots and short user notes.
• A multi-agent workflow where smaller subagents handle narrow tasks in parallel.

Inputs

• A required instruction or question in plain text that describes the task and the desired output.
• Optional image attachments for visual understanding, like screenshots, photos, charts, or scanned pages.
• If you attach images, use supported formats: PNG, JPG/JPEG, WEBP, or a non-animated GIF.
• Keep images readable. Tiny text, blur, glare, or heavy compression can reduce accuracy.
• Avoid watermarks, logos, and any NSFW image content.
• On Wiro, you can attach multiple images per run. This integration supports up to 10 images.
• Optional system-level instructions that set the assistant’s role, tone, and hard rules.
• Optional chat identifiers that let Wiro keep conversation history across multiple turns.
• Optional reasoning depth control. Choose from none, low, medium, high, or xhigh.
• Optional response length control. Choose low, medium, or high verbosity.
• Optional web search toggle. Enable it when your prompt needs fresh public sources.

Outputs

The model returns a text response. It can be plain answers, step-by-step guidance, summaries, or code blocks. If you request structured output, it can produce strict JSON for downstream parsing.

Limitations

• Output is text only. It does not generate images, audio, or video.
• It supports image understanding, but not audio or video understanding.
• The model has a fixed knowledge cutoff. It may miss events after August 31, 2025.
• The maximum context window is 400,000 tokens, and the maximum output is 128,000 tokens.
• Image quality matters. Low-resolution scans, skewed photos, and dense spreadsheets can cause mistakes.
• Complex charts and tiny UI labels may need higher reasoning effort and clearer images.
• Some platforms or gateways may enforce a smaller context window than the model’s maximum.

Safety & compliance

GPT-5.4 mini is safety-trained and will refuse disallowed requests, including harmful instructions and sexual content involving minors. Don’t use it to generate or enable wrongdoing. Don’t upload sensitive personal data unless you have a clear legal basis and user consent. If you enable web search, treat third-party content as untrusted and verify important claims before acting on them.