OpenAI launches GPT-4.1, and it is focused on performance and code
According to OpenAI, GPT-4.1 is 40% faster and 26% cheaper than the previous generation, the GPT-4o.
OpenAI has a habit of dropping models faster than most people can keep track of. But its latest release model family, GPT-4.1, is quietly one of the company’s most strategic updates yet. And while the naming still sounds like something out of a software patch log, this one is focused tightly and intentionally on code for developers.
Unveiled during a livestream Monday, the GPT-4.1 model family comes in three versions: the standard model, GPT-4.1 Mini, and GPT-4.1 Nano. Each one is designed to excel at writing code, following instructions, and working with massive inputs.
All three models can now process up to one million tokens in a single context window. That means they can take in the equivalent of an entire codebase, a full technical manual, or a multi-part prompt with room to spare. Compared to GPT-4o’s 128,000-token limit, this is a major leap in how much information the model can reason over at once.
OpenAI says GPT-4.1 is not just faster (by 40%) and cheaper (26% less than GPT-4o) but also more accurate in spotting relevant code and ignoring irrelevant filler. The company says it even beats GPT-4.5, the company’s most powerful model launched in February, in some key areas.
Emmanuel Oyedeji
According to OpenAI’s own benchmarks, scored GPT-4.1 up to 54.6% on SWE-bench Verified. Although that’s just behind Google’s Gemini 2.5 Pro (63.8%) and Anthropic’s Claude 3.7 Sonnet (62.3%), it costs way less. At just $2 per million input tokens and $8 per million output, GPT-4.1 undercuts its predecessors by 26%. And while the full-size model is the most powerful, Mini and Nano are even cheaper.
In testing, the model showed strong gains in understanding repositories, running unit tests, and generating functional, clean code. During the livestream, OpenAI demonstrated GPT-4.1 building a flashcard app, navigating through file structures, and writing code that compiled without issue.
The Big Picture
The 4.1 launch isn’t just about incremental improvement. It signals a shift. The long-term goal is to create what it calls an “agentic software engineer” — a model that can handle the full software development lifecycle, from writing and testing code to debugging and documenting it. GPT-4.1 is a step in that direction.
To make room, OpenAI is phasing out the old GPT-4 in ChatGPT by April 30 and sunsetting GPT-4.5 preview in the API this July. GPT-4.1 is set to be the new workhorse.
Meanwhile, the AI arms race is heating up with competitors such as Google, Anthropic, and DeepSeek are all ramping up to build top-tier developer tools. OpenAI has now made its move for faster, cheaper, and smarter models optimised for real-world coding and if GPT-4.1 is any sign, the days of AI as just a chatbot are numbered.
Wisdom Okogho
