> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bytespike.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# gpt-5-4-mini

> OpenAI's small-mid GPT — production-grade reasoning at a fraction of gpt-5-4's price. Use it for classification, routing, structured extraction, and short-prompt agent loops.

`gpt-5-4-mini` is the small-mid GPT-5. It keeps the GPT-5 reasoning capability but at roughly **one-third** the per-token cost of `gpt-5-4`. The right default for high-volume traffic where you want better-than-Haiku quality without paying flagship rates.

**Pricing:** $0.75 / 1M input, $4.50 / 1M output, \$0.075 / 1M cache read — see the [rate card](/pricing).

## Protocols

| Protocol                | Path                                                |
| ----------------------- | --------------------------------------------------- |
| OpenAI Chat Completions | `POST https://llm.bytespike.ai/v1/chat/completions` |
| OpenAI Responses        | `POST https://llm.bytespike.ai/v1/responses`        |

## Quickstart

<CodeGroup>
  ```bash cURL theme={null}
  curl https://llm.bytespike.ai/v1/chat/completions \
    -H "Authorization: Bearer $BYTESPIKE_API_KEY" \
    -H "content-type: application/json" \
    -d '{
      "model": "gpt-5-4-mini",
      "reasoning_effort": "low",
      "messages": [
        { "role": "user", "content": "Classify: refund / billing / technical / other.\n\nMy invoice charged twice." }
      ]
    }'
  ```

  ```python Python (Chat Completions) theme={null}
  from openai import OpenAI

  client = OpenAI(
      base_url="https://llm.bytespike.ai/v1",
      api_key=os.environ["BYTESPIKE_API_KEY"],
  )

  resp = client.chat.completions.create(
      model="gpt-5-4-mini",
      reasoning_effort="low",
      messages=[{"role": "user", "content": "Classify this support ticket."}],
  )

  print(resp.choices[0].message.content)
  ```
</CodeGroup>

## Capabilities

| Capability         | Supported            |
| ------------------ | -------------------- |
| Chat Completions   | ✅                    |
| Responses API      | ✅                    |
| Streaming (SSE)    | ✅                    |
| Vision             | ✅                    |
| Tool use           | ✅ parallel           |
| JSON mode          | ✅                    |
| Structured outputs | ✅                    |
| Reasoning effort   | ✅ (`low` / `medium`) |
| Web search         | —                    |
| Context window     | 128K tokens          |

## When to use

* **High-volume classification, routing, structured extraction** — anywhere you'd use Haiku for cost but want GPT-flavored reasoning.
* **Short-prompt agent loops** — tool-use chains where each step is small and the gate is throughput.
* **Cheaper Codex-style** via Responses API at `reasoning_effort: "low"`.

When **not** to use:

* Hard reasoning that benefits from `reasoning_effort: "high"` — go to `gpt-5-4` or `gpt-5-5`.
* Web search needed — only `gpt-5-4` and up have it.

## Next

* [gpt-5-4](/models/gpt-5-4) — production workhorse
* [gpt-5-5](/models/gpt-5-5) — current GPT-5 flagship