OpenAI · all models ›

OGPT-5 MiniAPI

GPT-5 Mini is a compact version of GPT-5, designed to handle lighter-weight reasoning tasks.

400KContext window · tokens
$0.25Input price · per M tokens
$2Output price · per M tokens
OpenAIProvider

Prices update automatically — checked daily against provider list prices.

See open-source alternatives → Compare all model prices

Benchmarks & performance

Independent benchmark scores for GPT-5 Mini, measured by Artificial Analysis. Higher is better. Measurement mode: medium.

Intelligence index31.6
Math index85
GPQA80.3%
MMLU-Pro82.8%
Humanity's Last Exam15.9%
Long Context Reasoning69.7%
LiveCodeBench69.2%
SciCode41%
AIME 202585%
IFBench71.2%
τ²-Bench71.1%
Terminal-Bench Hard28.8%
💰 Blended price$0.688 / 1M tokens
📈 Value45.9 intelligence points per $
vs. models measured heretop 46%
Scores higher than 54% of the 273 models measured by Artificial Analysis and tracked here.
Models at this level cost $1.08 per 1M tokens (median of 20) — this one costs $0.69.
Cheaper and better on this index: GLM 5.3 Flash · Gemini 3.7 Flash (batch) · GPT-5.6 Luna and 26 more
Benchmark data by Artificial Analysis

About this model

GPT-5 Mini is a commercial AI model by OpenAI. The specifications below are tracked automatically: pricing is refreshed daily from public list prices, so the numbers on this page reflect the current cost of using the model through its API.

Frequently asked questions

What is GPT-5 Mini?

GPT-5 Mini is an AI language model from OpenAI. It is a proprietary model, available through an API. It scores 31.6 on the Artificial Analysis intelligence index.

Is GPT-5 Mini free?

GPT-5 Mini is not free: it costs $0.25 per million input tokens and $2 per million output tokens. Open-weight alternatives can be self-hosted at no per-token cost.

What is GPT-5 Mini good at?

Independent benchmarks from Artificial Analysis give it GPQA 80.3%, MMLU-Pro 82.8%, Humanity's Last Exam 15.9%, Long Context Reasoning 69.7%, LiveCodeBench 69.2%, SciCode 41%, AIME 2025 85%, IFBench 71.2%, τ²-Bench 71.1%, Terminal-Bench Hard 28.8%. It is particularly used for mathematical reasoning.

Related models

o1-proOpenAIo1-pro (batch)OpenAIGPT-5.5 ProOpenAIGPT-5.4 ProOpenAIGPT-5.2 ProOpenAIGPT-5 ProOpenAI