GPT-4o mini

// Overview

GPT-4o mini is OpenAI's small, low-cost model introduced on July 18, 2024 as the successor to GPT-3.5 Turbo. It launched at $0.15 per million input tokens and $0.60 per million output tokens — which OpenAI said made it more than 60% cheaper than GPT-3.5 Turbo — and immediately became the default model for free-tier ChatGPT users while also being available in the API.

Despite the low price, GPT-4o mini posted strong scores for its tier: OpenAI reported 82.0% on MMLU, 87.0% on the MGSM math benchmark, and 87.2% on HumanEval coding, beating rival small models of the era such as Google's Gemini 1.5 Flash and Anthropic's Claude 3 Haiku on those tests. It carries a 128K-token context window, returns up to 16,384 output tokens per request, and has an October 2023 knowledge cutoff.

At launch GPT-4o mini supported text and vision (image) inputs with text output; OpenAI said video and audio support could come later via the API. For two years it was OpenAI's go-to model for high-volume, latency-sensitive, and cost-sensitive tasks like classification, extraction, summarization, and chat assistants. It remains available through the OpenAI API even after the older 4o-class models were retired from the ChatGPT interface during OpenAI's 2026 legacy-model cleanup.

// Benchmarks

1. MMLU (5-shot)82%
2. MGSM (math, 0-shot CoT)87%
3. HumanEval (coding)87.2%
4. MMMU (multimodal reasoning)59.4%
5. MATH70.2%
6. GPQA40.2%
7. DROP (F1)79.7%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Launch pricing as announced by OpenAI on July 18, 2024; OpenAI stated this was more than 60% cheaper than GPT-3.5 Turbo. Batch API and cached-input discounts apply separately.

Pricing source ↗

// Strengths

• Very low price — $0.15 input / $0.60 output per million tokens, far cheaper than GPT-4o and a steep cut from GPT-3.5 Turbo
• Fast, low-latency responses suited to high-throughput and real-time workloads
• Strong reasoning and coding for its size class (MMLU 82%, HumanEval 87.2%) — outscored Gemini 1.5 Flash and Claude 3 Haiku at launch
• 128K-token context window — long enough to process large documents and long conversations
• Native vision: accepts image inputs alongside text
• Supports function/tool calling, structured outputs, and fine-tuning

// Best for

• High-volume text classification, tagging, and content moderation where cost per call matters
• Information extraction and structured-data parsing from documents and emails
• Summarization of long documents, transcripts, and chat logs within the 128K window
• Customer-facing chatbots and assistants needing fast, cheap responses
• Drafting, rewriting, and routine coding help where frontier-model quality isn't required
• Function/tool-calling agents and pipelines that fan out many small LLM calls

// How to access

// GPT Mini — every version

The full lineage of the GPT Mini line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.