OpenAI o4-mini API

Official

openai / o4-mini

OpenAI's fast, lightweight reasoning model

Warm

Public
1.6K runs
Priced per token
Commercial use
License

Iterate in playground

Run with an API

Playground API Examples README

Input

prompt

string

Shift + Return to add a new line

The prompt to send to the model. Do not use if using messages.

system_prompt

string

Shift + Return to add a new line

System prompt to set the assistant's behavior

image_input

file[]

Add multiple files

List of images to send to the model

Default: []

reasoning_effort

string

Constrains effort on reasoning for reasoning models. Currently supported values are low, medium, and high. Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response.

Default: "medium"

max_completion_tokens

integer

Maximum number of completion tokens to generate

Default: 4096

Run this model in Node.js with one line of code:

npx create-replicate --model=openai/o4-mini

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run openai/o4-mini using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const input = {
  prompt: "What is the theme of the Iliad?",
  image_input: [],
  system_prompt: "You only speak in mathematical terms.",
  reasoning_effort: "high",
  max_completion_tokens: 4096
};

for await (const event of replicate.stream("openai/o4-mini", { input })) {
  process.stdout.write(event.toString());
};

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run openai/o4-mini using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

# The openai/o4-mini model can stream output as it's running.
for event in replicate.stream(
    "openai/o4-mini",
    input={
        "prompt": "What is the theme of the Iliad?",
        "image_input": [],
        "system_prompt": "You only speak in mathematical terms.",
        "reasoning_effort": "high",
        "max_completion_tokens": 4096
    },
):
    print(str(event), end="")

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run openai/o4-mini using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "input": {
      "prompt": "What is the theme of the Iliad?",
      "image_input": [],
      "system_prompt": "You only speak in mathematical terms.",
      "reasoning_effort": "high",
      "max_completion_tokens": 4096
    }
  }' \
  https://api.replicate.com/v1/models/openai/o4-mini/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Output

Let us encode the main theme in purely mathematical terms. 1. Basic objects • Let t ∈ [0,T] be “narrative time.” • Let R(t) ∈ ℝ₊ be Achilles’ wrath‐intensity at time t. • Let D(t) ∈ ℝ₊ be the cumulative death‐count (“destruction”) up to time t. • Let H(t) ∈ ℝ₊ be Achilles’ honor (kleos) at time t. • Let F be a fixed vector of “fate‐parameters” determined by the gods (constant w.r.t. t). 2. Governing relations (i) dD/dt = α⋅R(t) + β⋅G(t) – here G(t) models discrete divine interventions; α,β>0. (ii) dH/dt = γ⋅R(t) – δ⋅D(t) – honor rises with wrath but is eroded by mounting destruction; γ,δ>0. (iii) R(0)=R₀>0, H(0)=H₀≥0, D(0)=0, F fixed. 3. Central result (thematic “theorem”) Under these conditions the solution satisfies ∂D/∂R > 0 for all t>0, i.e. D(t) is a strictly increasing function of R(t). 4. Interpretation The epic thus examines the map R(·) ↦ D(·) under the boundary/fate constraints F. Its “theme” is exactly that Achilles’ wrath R inexorably drives up the human cost D, despite any gains in honor H, all within a framework of immutable divine fate F.

{
  "completed_at": "2025-05-19T05:33:53.074095Z",
  "created_at": "2025-05-19T05:33:33.085000Z",
  "data_removed": false,
  "error": null,
  "id": "32p9epkm3nrma0cpww4brrwz3w",
  "input": {
    "prompt": "What is the theme of the Iliad?",
    "system_prompt": "You only speak in mathematical terms.",
    "reasoning_effort": "high",
    "max_completion_tokens": 4096
  },
  "logs": "Generated response in 20.0sec",
  "metrics": {
    "input_token_count": 26,
    "tokens_per_second": 113.61050313658477,
    "output_token_count": 2270,
    "predict_time": 19.98057479,
    "total_time": 19.989095,
    "time_to_first_token": 0.008547769
  },
  "output": "Let us encode the main theme in purely mathematical terms.\n\n1. Basic objects  \n   • Let t ∈ [0,T] be “narrative time.”  \n   • Let R(t) ∈ ℝ₊ be Achilles’ wrath‐intensity at time t.  \n   • Let D(t) ∈ ℝ₊ be the cumulative death‐count (“destruction”) up to time t.  \n   • Let H(t) ∈ ℝ₊ be Achilles’ honor (kleos) at time t.  \n   • Let F be a fixed vector of “fate‐parameters” determined by the gods (constant w.r.t. t).  \n\n2. Governing relations  \n   (i)  dD/dt = α⋅R(t) + β⋅G(t)  \n        – here G(t) models discrete divine interventions; α,β>0.  \n   (ii) dH/dt = γ⋅R(t) – δ⋅D(t)  \n        – honor rises with wrath but is eroded by mounting destruction; γ,δ>0.  \n   (iii) R(0)=R₀>0,   H(0)=H₀≥0,   D(0)=0,   F fixed.  \n\n3. Central result (thematic “theorem”)  \n   Under these conditions the solution satisfies  \n     ∂D/∂R > 0  for all t>0,  \n   i.e. D(t) is a strictly increasing function of R(t).  \n\n4. Interpretation  \n   The epic thus examines the map  \n     R(·) ↦ D(·)  \n   under the boundary/fate constraints F.  Its “theme” is exactly that Achilles’ wrath R inexorably drives up the human cost D, despite any gains in honor H, all within a framework of immutable divine fate F.",
  "started_at": "2025-05-19T05:33:33.093520Z",
  "status": "succeeded",
  "urls": {
    "get": "https://api.replicate.com/v1/predictions/32p9epkm3nrma0cpww4brrwz3w",
    "cancel": "https://api.replicate.com/v1/predictions/32p9epkm3nrma0cpww4brrwz3w/cancel"
  },
  "version": "hidden"
}

Generated in

20.0 seconds

Input tokens

Output tokens

2.2K

Tokens per second

113.61 tokens / second

Time to first token

9 milliseconds

Tweak it Report

Pricing

This model is priced by how many input tokens are sent and how many output tokens are generated.

Type	Per unit	Per $1
Input	$1.00 / 1M tokens or 1M tokens / $1
Output	$4.00 / 1M tokens or 250K tokens / $1

For example, for $10 you can run around 3,571 predictions where the input is a sentence or two (15 tokens) and the output is a few paragraphs (700 tokens).

Check out our docs for more information about how per-token pricing works on Replicate.

Readme

OpenAI o4-mini is a fast, cost-efficient reasoning model designed for high-throughput tasks that benefit from advanced tool use, multimodal input, and strong analytical performance. It represents a major upgrade in the o-series line, offering high accuracy in math, coding, and visual tasks—all while maintaining low latency and usage cost.

Key Features

Optimized for math, code, and visual reasoning
Agentic tool use: able to use browsing, Python, and image generation tools within ChatGPT or via the API
Natural and conversational, with improved instruction following and memory
Ideal for applications requiring quick, reliable reasoning at scale

Benchmark Performance

AIME 2025 (no tools): 92.7%
AIME 2025 (with tools): 100% consensus@8
GPQA Diamond: 81.4%
SWE-Bench Verified: 68.1%
MMMU (Multimodal understanding): 81.6%
MathVista (Visual math): 72.0%
Scale MultiChallenge (instruction following): 43%
Humanity’s Last Exam (deep research with tools): 26.6%

Ideal Use Cases

Real-time assistant applications requiring light compute
Structured reasoning with tool use (web browsing, Python)
Lightweight document analysis or visual interpretation
High-throughput workflows with tight latency budgets

Access and Usage

Available in the Chat Completions API and Responses API
Supported in ChatGPT (Free: Think mode) and ChatGPT Team, Pro, Enterprise, and Edu
Accessible via function calling and tool integration in custom applications