What to know

  • OpenAI has released GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano, offering up to 1 million tokens of context.
  • The new models show up to 27% improvement in coding abilities over GPT-4.5 and are 26% more efficient than GPT-4o.
  • GPT-4.1 models are available only via API and will replace GPT-4.5, which will be discontinued by July 14, 2025.
  • These models are trained on data up to June 2024 and are designed for developers, with enhanced instruction-following and image understanding.

OpenAI has introduced its latest flagship artificial intelligence models, the GPT-4.1 family, which includes the main GPT-4.1 model, as well as the smaller GPT-4.1 Mini and GPT-4.1 Nano versions. These models are now available through OpenAI's API and are designed to serve a wide range of developer needs, from handling large-scale codebases to delivering fast, low-latency responses for simpler tasks.

The standout feature of the GPT-4.1 models is their ability to process up to 1 million tokens of context in a single call. This represents a substantial increase over previous models, such as GPT-4o, which supported up to 128,000 tokens. The expanded context window allows the models to handle entire code repositories, lengthy documents, and complex workflows without losing track of relevant information.

OpenAI reports that GPT-4.1 demonstrates a 21% improvement in coding abilities over GPT-4o and a 27% improvement over GPT-4.5. The models also consume 26% less computing power than GPT-4o, making them more efficient for developers and organizations. The new family is particularly effective at editing large files, following instructions, and understanding images, surpassing previous models on several benchmarks.

With this release, OpenAI is phasing out the GPT-4.5 Preview, which will be turned off in the API by July 14, 2025. The company states that GPT-4.1 offers similar or better performance at a lower cost and with reduced latency. The models are trained on data up to June 2024, ensuring they have the most recent knowledge available among OpenAI's offerings.

GPT-4.1, Mini, and Nano are available exclusively via the API and are not integrated into the ChatGPT interface. OpenAI has worked closely with the developer community to optimize these models for real-world coding and instruction-following tasks. The Nano version is the fastest and most cost-effective, suitable for classification and autocompletion, while the full-size model excels at handling complex, multi-turn agent workflows.

Overall, the GPT-4.1 family marks a significant step forward in AI capabilities for developers, offering improved performance, efficiency, and the ability to process much larger contexts than before.

Via: The Verge