OpenAI has unveiled a new suite of models under the GPT-4.1 family, consisting of GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. These models are designed to surpass their predecessors, GPT-4o and GPT-4o mini, in nearly every metric, offering enhanced performance and capabilities for developers. One of the standout features of GPT-4.1 models is their vastly increased context window, now reaching a million tokens—dramatically higher than the 128,000 tokens supported by GPT-4o. This expansion improves long-context comprehension and paves the way for more sophisticated use cases, such as extracting insights from lengthy documents or processing more intricate instructions.
In a move aimed at streamlining its offerings, OpenAI announced the deprecation of the GPT-4.5 Preview API, which will be fully discontinued by July 14, 2025. According to OpenAI, GPT-4.1 delivers equivalent or superior performance across many functionalities while being more cost-effective and faster than GPT-4.5, making it a more practical choice for developers. The increased output token limits, now extending up to 32,767 tokens, further reinforce the model’s suitability for handling complex and resource-intensive tasks.
However, OpenAI has stated that GPT-4.1 will only be available through the API and will not be integrated into ChatGPT. While this may disappoint some users, OpenAI emphasized that many of the GPT-4.1 enhancements have already been incorporated into the latest GPT-4o version, with additional improvements planned for future releases. The decision to focus on API access also highlights the model’s orientation towards developers and businesses that require customized solutions.
Among the most notable advancements are in coding capabilities. OpenAI reports a 21.4% improvement in coding performance on the SWE-bench when compared to GPT-4o, making GPT-4.1 models a powerful tool for software development tasks. The GPT-4.1 mini model, in particular, has shown significant strides in small model performance, even outpacing GPT-4o in various benchmarks. For tasks that demand rapid responses with minimal cost, the GPT-4.1 nano model excels, offering exceptional performance for tasks such as classification, autocompletion, and other lightweight operations. These enhancements are set to allow developers to create more efficient and reliable agents capable of performing complex, real-world tasks with less manual intervention.