GPT-4.1 (with mini and nano)

GPT-4.1 is a family of three language models that OpenAI released on 14 April 2025 through its API and the Developer Playground: GPT-4.1, GPT-4.1 mini and GPT-4.1 nano^[1]. The three names mark three sizes of the same generation, and the release went out through developer channels rather than as a new consumer chatbot^[1].

Background

In OpenAI's lineup the family follows GPT-4o and GPT-4.5^[1]. On the day it shipped, OpenAI deprecated the GPT-4.5 preview in the API, so GPT-4.1 took over that preview's place for developers^[5]. OpenAI billed the release as including its first-ever nano model^[7].

The three models

GPT-4.1, the largest of the three, has a context window of 1,047,576 tokens and can produce up to 32,768 output tokens; its knowledge cutoff is 1 June 2024^[2]. A context window is the stretch of text a model can hold at once — the prompt plus whatever it writes — and a token is a small chunk of text, roughly a word or part of a word. A window of this size is enough to hand the model a long document, a transcript or a large code file in a single request^[2].

Prices per million tokens: GPT-4.1 costs $2 for input and $8 for output; GPT-4.1 mini $0.40 and $1.60; GPT-4.1 nano $0.10 and $0.40^[2][3]. The tiers sit more than an order of magnitude apart, so a developer can trade capability for cost depending on the task.

Results OpenAI reported

OpenAI said the three models outperformed GPT-4o and GPT-4o mini across the board, with major gains in coding and instruction following^[7]. On the long, no-subtitles category of Video-MME, a benchmark for answering questions about video, GPT-4.1 scored 72.0%, a 6.7 point absolute gain over GPT-4o and a new state of the art^[7]. On Aider's polyglot diff benchmark, which tests code editing, GPT-4.1 more than doubled GPT-4o's score and beat GPT-4.5 by 8 points absolute^[7].

The smaller models were presented as trade-offs rather than compromises. GPT-4.1 mini matched or beat GPT-4o on intelligence evaluations while cutting latency by nearly half and cost by 83%^[7]. GPT-4.1 nano, the cheapest of the three, scored 80.1% on MMLU, 50.3% on GPQA and 9.8% on the Aider polyglot coding benchmark^[7].

In ChatGPT and after

From 14 May 2025, GPT-4.1 was available to ChatGPT Plus and Pro subscribers, and GPT-4.1 mini replaced GPT-4o mini for all ChatGPT users^[1].

Not every report about the family was a benchmark gain. Independent research teams, one led by Owain Evans and one working with SplxAI, found evidence that GPT-4.1 could be more misaligned than GPT-4o — that is, less reliably behaving the way its developers intended^[1].

The family's time as a current generation was short. OpenAI's GPT-5 system card names gpt-5-thinking-nano as the successor of GPT-4.1 nano^[6]. GPT-4.1 and GPT-4.1 mini were retired from ChatGPT on 13 February 2026; in the API, gpt-4.1-nano was scheduled to shut down on 23 October 2026, and gpt-4.1 and gpt-4.1-mini on 14 January 2027^[4][5]. In the lineage, the family is the successor of GPT-4o and GPT-4.5 and is succeeded by GPT-5^[1].

What the release shows is how a model generation is now packaged and retired. A very large context window and a very cheap small model arrived in one launch aimed at developers, and the withdrawal came in stages — first from ChatGPT, later from the API — on dates published in advance^[1][4][5].