Introducing GPT-4.1

GPT-4.1 is the latest advancement in OpenAI's series of generative pre-trained transformers, offering substantial improvements across various domains. This new family of models is designed to enhance performance in coding, instruction following, and long-context comprehension, making it a versatile tool for developers worldwide.

Key Features

• Enhanced Coding Capabilities: GPT-4.1 scores 54.6% on SWE-bench, a 21.4% improvement over its predecessor, making it a leading model for coding tasks.
• Improved Instruction Following: Achieves a 38.3% score on Scale's MultiChallenge benchmark, showcasing a 10.5% increase in instruction-following ability.
• Long-Context Understanding: Supports up to 1 million tokens of context, setting a new state-of-the-art result on the Video-MME benchmark with a 72.0% score.
• Cost and Latency Efficiency: Offers exceptional performance at a lower cost, with the GPT-4.1 mini reducing latency by nearly half and cost by 83% compared to previous models.
• Nano Model Availability: The GPT-4.1 nano model is optimized for tasks requiring low latency, such as classification or autocompletion.

Use Cases

GPT-4.1 is ideal for developers looking to build advanced AI applications. It is particularly beneficial for:

• Software engineering tasks
• Extracting insights from large documents
• Resolving customer requests with minimal intervention
• Developing intelligent agents that can perform tasks independently

Unique Selling Points

GPT-4.1 stands out with its ability to handle extensive context windows and its superior performance in coding and instruction-following tasks. Its nano model offers unmatched speed and efficiency for specific applications.

Technical Details

• Platforms: Available exclusively via the OpenAI API
• Integrations: Compatible with the Responses API for building intelligent agents
• Knowledge Cutoff: Updated to June 2024

"GPT-4.1 models push performance forward at every point on the latency curve, offering exceptional performance at a lower cost."