OpenAI o4-mini

What it is

o4-mini is a small reasoning model released by OpenAI on 16 April 2025 alongside o3 ^[1]. A reasoning model is trained to work through a problem in explicit intermediate steps before answering, spending more computation at inference time than a model that simply continues text. o4-mini was the smaller, cheaper member of that family, and OpenAI described it as a model optimised for fast, cost-efficient reasoning ^[6].

Specifications and price

Through the API, o4-mini offers a 200,000-token context window and can produce up to 100,000 output tokens; its knowledge cutoff is 1 June 2024, and it accepts both text and images as input ^[2]. A token is the unit of text a model reads and writes, roughly a short word or a fragment of one; the context window therefore sets how much material can be in front of the model at once, and the output limit sets how long a single answer can be. Pricing was $1.10 per million input tokens and $4.40 per million output tokens ^[2].

Benchmarks

OpenAI described o4-mini as its best-performing benchmarked model on AIME 2024 and AIME 2025 ^[6]. AIME is a high-school mathematics competition whose problems are widely used to test multi-step quantitative reasoning, and pass@1 measures how often a model's first submitted answer is correct. Given access to a Python interpreter — that is, allowed to write and run code as part of solving — o4-mini reached 99.5% pass@1 on AIME 2025 ^[6]. Calling a tool in the middle of reasoning lets the model hand off arithmetic and programmatic checks instead of carrying them out internally.

Usage and evaluation

OpenAI positioned o4-mini for volume: it supported significantly higher usage limits than o3, which made it a high-volume, high-throughput reasoning option ^[6]. In expert evaluations it outperformed o3-mini on non-STEM tasks and on domains such as data science ^[6]. Because o3-mini was the previous small reasoning model, that comparison describes movement within the same size class rather than a jump between weight classes.

In ChatGPT

In ChatGPT, Plus, Pro and Team users saw both o4-mini and o4-mini-high in the model selector, while free users could try o4-mini by selecting "Think" ^[6]. The selector thus exposed more than one variant of the same model, and reasoning was also reachable by free users through a dedicated entry point.

Retirement

o4-mini's life under that name was short. OpenAI's GPT-5 system card names gpt-5-thinking-mini as its successor ^[5], and o4-mini was retired from ChatGPT on 13 February 2026 ^[4]. The API model was scheduled to shut down on 23 October 2026, with GPT-5 mini as its successor ^[2][3]. A base model, a "-high" variant, and then consolidation into a successor family is the pattern many model line-ups followed in this period.