Reading · Benjamin Crozat's blog posts

What this GPT-5.4 mini guide covers

OpenAI introduced the GPT-5.4 family on March 5, 2026 in its GPT-5.4 launch post. The current GPT-5.4 mini model page describes mini as OpenAI’s strongest mini model yet for coding, computer use, and subagents.

That makes GPT-5.4 mini the practical middle ground in the GPT-5.4 family. The model page says it brings the strengths of GPT-5.4 to a faster, more efficient model designed for high-volume workloads.

If you are starting fresh and want the flagship option first, read my GPT-5.4 API guide. If you only care about the cheapest branch, compare this with GPT-5.4 nano.

By the end, you will have:

a first successful GPT-5.4 mini Responses API call
a structured support-triage workflow that returns strict JSON

Get your API key ready

You need an OpenAI account, a funded API project, and an API key from the API keys page. Keys are shown once, so save yours right away.

Then export it in your terminal.

macOS and Linux:

export OPENAI_API_KEY="sk-..."

Windows Command Prompt:

setx OPENAI_API_KEY "sk-..."

If you use setx, open a new terminal before testing the key.

Send your first GPT-5.4 mini request

GPT-5.4 mini uses the same Responses API shape as the rest of the GPT-5.4 family, so the first request stays simple.

This exact request is a good first check:

curl -s https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4-mini-2026-03-17",
    "input": [
      {
        "role": "user",
        "content": [
          { "type": "input_text", "text": "Say hello in one short sentence." }
        ]
      }
    ],
    "text": {
      "verbosity": "low"
    },
    "max_output_tokens": 80
  }'

If your key and billing are set up correctly, that gives you a quick way to verify the endpoint and model selection are working.

The model page also lists the parts that matter for real apps:

400,000 context window
128,000 max output tokens
text and image input
structured outputs
function calling
web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search

Build something useful: support triage

Now let us turn the first successful call into something you could actually ship.

Imagine your app receives this support message:

Hi, I was billed twice for my Pro plan today. Please refund the extra charge.

You want GPT-5.4 mini to do four things in one pass:

classify the issue
set a priority
decide whether a human should step in
draft a safe reply

That is a good fit for this model because the output is bounded, the prompt is precise, and the result is easy to validate.

Return strict JSON with a schema

curl -s https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4-mini-2026-03-17",
    "instructions": "You triage support messages for a SaaS app. Be cautious. Do not promise actions the billing team has not confirmed. Keep reply_draft to 2 short sentences.",
    "input": [
      {
        "role": "user",
        "content": [
          {
            "type": "input_text",
            "text": "Hi, I was billed twice for my Pro plan today. Please refund the extra charge."
          }
        ]
      }
    ],
    "text": {
      "verbosity": "low",
      "format": {
        "type": "json_schema",
        "name": "support_triage",
        "schema": {
          "type": "object",
          "properties": {
            "category": {
              "type": "string",
              "enum": ["billing", "bug", "account", "feature_request", "other"]
            },
            "priority": {
              "type": "string",
              "enum": ["low", "medium", "high"]
            },
            "needs_human": {
              "type": "boolean"
            },
            "reply_draft": {
              "type": "string"
            }
          },
          "required": ["category", "priority", "needs_human", "reply_draft"],
          "additionalProperties": false
        },
        "strict": true
      }
    },
    "max_output_tokens": 220
  }'

If everything is wired correctly, you should get JSON shaped like this:

{
  "category": "billing",
  "priority": "high",
  "needs_human": true,
  "reply_draft": "Sorry about the duplicate charge. I have flagged this to our billing team to review the transaction and follow up with you."
}

That is immediately useful in an app:

category can route the ticket
priority can change queue order
needs_human can trigger escalation
reply_draft can prefill the first response

If you want to move this pattern into PHP next, my guide on using OpenAI’s API in PHP with openai-php/client picks up right there.

What changes with GPT-5.4 mini

GPT-5.4 mini sits in a sweet spot.

The model page backs up a clear value proposition: a faster, more efficient model for high-volume workloads. The pricing page puts it at $0.75 input and $4.50 output per 1M tokens, while the same comparison area shows GPT-5.4 at $2.50 input and $15 output per 1M tokens. That makes GPT-5.4 mini much cheaper than the full model while still staying in the newer GPT-5.4 family.

So if your task is precise and repeatable, mini often makes more sense than the flagship model. If the work is even narrower and cost is the main constraint, GPT-5.4 nano is the next model I would compare.

How I would use GPT-5.4 mini

I would reach for GPT-5.4 mini when:

the prompt is clear
the output shape is predictable
the task needs to be fast and affordable
the job is a coding assistant subtask, a computer-use step, or a structured classification workflow

I would not use it as a blanket default for messy reasoning. If the task is ambiguous, long, or coordination-heavy, full GPT-5.4 still has the advantage.

Common mistakes with GPT-5.4 mini

1. Treating it like the flagship by default

Mini is excellent when the job is bounded. It is not the right starting point for every hard problem.

2. Using vague prompts

This model shines when you tell it exactly what to classify, extract, or decide.

3. Skipping the cheaper or larger sibling when the fit is obvious

If you only need very fast, very cheap classification, GPT-5.4 nano may be enough. If the task needs more breadth or judgment, GPT-5.4 is the better upgrade path.

If GPT-5.4 mini looks close to what you need, these are the next reads I would keep open:

GPT-5.4 mini API quick start with a real workflow