OpenAI Launches GPT-4.1: Next-Gen AI Models for Superior Programming and Contextual Understanding

OpenAI has unveiled a new lineup of AI models: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. These models excel at programming tasks and executing instructions efficiently.

The neural networks are accessible via an API and feature multimodal capabilities with a contextual window of up to a million tokens, allowing them to process approximately 750,000 words in a single instance.

This latest release marks a significant advancement towards developing AI models capable of creating entire applications while ensuring high quality, error testing, and comprehensive documentation writing.

In benchmark evaluations, GPT-4.1 outperforms its predecessors, GPT-4o and GPT-4o mini, particularly in programming assessments like SWE-bench. The mini and nano versions are designed to be faster and more efficient, albeit with slightly reduced accuracy.

OpenAI highlighted that GPT-4.1 nano is the fastest and most cost-effective model to date. The pricing structure is as follows:

According to internal tests, GPT-4.1 achieved scores ranging from 52% to 54.6% on the SWE-bench Verified benchmark and 72% in the Video-MME evaluation that measures the ability to understand the context of non-captioned long-form videos.

OpenAI announced that access to GPT-4.5—their largest AI model—via API will cease on July 14. Developers will need to transition to another solution offered by the company, with GPT-4.1 being promoted as the preferred alternative.

«[GPT-4.1] provides comparable or even enhanced performance relative to GPT-4.5 in key areas, at a significantly lower cost,» emphasized the startup.

OpenAI charges $75 for every million tokens processed by GPT-4.5 and $150 for each million tokens generated, making it 30 times more expensive for input and 15 times more for output compared to GPT-4o.

However, GPT-4.5 will remain available within ChatGPT for users with paid subscriptions.

As a reminder, in April, GPT-4.5 successfully passed a three-party Turing test.