OpenAI GPT-4.1

GPT-4.1 is a family of transformer-based large language models designed with a strong developer focus, initially released exclusively through API access. It introduced three variants—a standard, mini, and nano model—to serve different performance and efficiency needs. The model's defining architectural capability is its support for up to one million tokens of context, dramatically expanding the ability to work with very long documents, extensive codebases, or multi-document analysis tasks. The refresh of the knowledge cutoff to June 2024 brought the model's world understanding closer to the present day, while improved long-context comprehension allows the model to reason more reliably across large contexts without losing track of earlier information.

The model was trained with an emphasis on real-world utility, and the benchmark results reflect targeted improvements in areas developers care about most. On SWE-bench Verified, a rigorous coding benchmark, GPT-4.1 achieved 54.6%—a 21.4 percentage point leap over its predecessor, positioning it as a leading choice for code generation and software engineering tasks. Instruction-following performance saw similar gains, with the Scale MultiChallenge benchmark showing a 10.5 point improvement over GPT-4o, meaning the model better adheres to complex, multi-step directives. The model also set a new state-of-the-art result on the Video-MME multimodal benchmark for long-context understanding, scoring 72.0% on the long, no-subtitles category. These strengths make GPT-4.1 particularly well-suited for developer workflows, automated agents, and applications requiring reliable instruction adherence across extended contexts.