GPT-4.1 is part of OpenAI's refreshed GPT family, launched on April 14, 2025, alongside smaller mini and nano variants designed to push the family further on coding, instruction following, and the use of very long inputs. The headline change is a context window that scales to roughly one million tokens, paired with a refreshed knowledge cutoff reaching June 2024, allowing the model to draw on recent information while still handling document- and codebase-scale prompts in a single pass. This positions GPT-4.1 as a workhorse for tasks where a model needs to keep large amounts of material in working memory rather than summarize them away.

In practice, GPT-4.1 shows its strongest gains on the most demanding developer- and assistant-style tasks. On the SWE-bench Verified coding benchmark it reaches 54.6%, a 21.4 percentage-point absolute jump over GPT-4o and 26.6 points over GPT-4.5, and on Scale's MultiChallenge it lifts instruction-following performance to 38.3%, up 10.5 points over GPT-4o. It also sets a new high on the long, no-subtitles split of Video-MME with 72.0%, a 6.7-point absolute improvement that reflects how much better it can reason over hour-long video transcripts and other very long multimodal inputs. These results make GPT-4.1 a natural fit for code generation and refactoring, multi-step agents that must respect detailed instructions, and applications such as enterprise document analysis or video understanding where million-token context is an advantage rather than a checkbox.