On 14 April 2025, OpenAIunveiled its latest suite of language models: GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano. These models represent significant advancements in artificial intelligence, offering improved coding proficiency, enhanced instruction-following abilities, and the capacity to process extensive contexts. Designed with developers in mind, the GPT-4.1 series aims to provide more efficient and cost-effective tools for a variety of applications.
GPT-4.1 (and -mini and -nano) are now available in the API!
these models are great at coding, instruction following, and long context (1 million tokens).
benchmarks are strong, but we focused on real-world utility, and developers seem very happy.
GPT-4.1 demonstrates notable improvements in coding tasks, achieving a 54.6% score on the SWE-bench Verified benchmark of a 21.4 percentage point increase over GPT-4o and a 26.6 point rise over GPT-4.5. The model has been fine-tuned to reduce unnecessary edits, adhere to specified formats, and maintain consistent tool usage, thereby enhancing its utility in real-world software engineering scenarios.
Expanded Contextual Understanding
A significant advancement in the GPT-4.1 series is its ability to process up to one million tokens of context, a substantial increase from the 128,000-token limit of previous models. This expanded context window enables the models to handle large datasets and complex documents more effectively, benefiting applications in legal analysis, coding, customer support, and more.
Improved Instruction Following
GPT-4.1 exhibits enhanced capabilities in following complex instructions, scoring 38.3% on Scale’s MultiChallenge benchmark, 10.5 percentage point improvement over GPT-4o. This improvement facilitates the development of more reliable AI agents capable of executing intricate tasks with greater accuracy.
Announcing GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano in the API.
TL;DR: Major improvements on coding, instruction following, and long context. 💥
00:00 Intro 02:18 Coding 04:53 Instruction following 06:58 Long context 10:22 Demos, pricing, and availability 20:00 @windsurf_aipic.twitter.com/vviVb1pRqV
The GPT-4.1 series comprises three distinct models:
GPT-4.1: The flagship model offering the highest performance across tasks.
GPT-4.1 Mini: A cost-effective version that balances performance and affordability, reducing latency by nearly half and cost by 80% compared to GPT-4o.
GPT-4.1 Nano: The fastest and most affordable model, ideal for tasks like classification or autocompletion, scoring 80.1% on MMLU and 50.3% on GPQA.
These models are accessible exclusively via OpenAI’s API and are also available through platforms like GitHub Copilot and Azure OpenAI Service.
Pricing Structure
OpenAI has structured the pricing of the GPT-4.1 models to accommodate various user needs:
GPT-4.1: $2.00 per million input tokens and $8.00 per million output tokens.
GPT-4.1 Mini: $0.40 per million input tokens and $1.60 per million output tokens.
GPT-4.1 Nano: $0.10 per million input tokens and $0.40 per million output tokens.
Additionally, OpenAI offers a 75% discount for cached inputs and a 50% discount for batch processing, further enhancing cost efficiency for developers.
Industry Context and Future Outlook
The release of the GPT-4.1 series comes amid increasing competition in the AI sector, with companies like Google and Anthropic also advancing their AI capabilities. OpenAI’s focus on practical utility, cost efficiency, and developer feedback positions it to maintain a strong presence in the market. Looking ahead, OpenAI plans to release an open-weight model in the summer, further supporting developers in customising AI models for specific applications.
Conclusion
OpenAI’s GPT-4.1 series represents a significant step forward in AI model development, offering enhanced coding abilities, improved instruction following, and expanded context comprehension. By providing a range of models tailored to different performance and cost requirements, OpenAI continues to support the evolving needs of developers and organisations in the AI landscape.
