OpenAI has released GPT-4.1, and new ‘mini’ and ‘nano’ models, with significant performance improvements, especially in coding, following instructions, and handling long contexts. The new line of products targets greater cost-effectiveness, reduced latency, and greater intelligence for mass-scale AI application such as code generation and advanced analysis.
As a production-ready step ahead of earlier versions (such as internal work potentially branded GPT-4.5), GPT-4.1 brings significant enhancements:
Coding: Accuracy rises to 54.6% on SWE-bench Verified (up from 33% for GPT-4o). Early customers experience considerably shorter development times due to improved use of tools and fewer bugs.
Instruction Following: GPT-4.1 is more conformant, particularly with difficult prompts, with improved test performance such as Scale AI’s MultiChallenge (38%) and greatly improved understanding and query accuracy for clients such as Blue J (53% improvement) and Hex (doubled accuracy).
Also Read: Zetech University Unveils Eco-friendly solar tuk-tuk to Make Transportation Cheaper.
All GPT-4.1 models process 1 million tokens, making large document or codebase inspection more robust, as supported by benchmarks and customers such as Thomson Reuters and Carlyle with improved data extractions and cross-referencing.
The new versions are also more efficient and quicker. GPT-4.1 mini cuts GPT-4o’s latency in half, while nano provides sub-five-second responses for lightweight tasks. GPT-4.1 is, on average, 26% cheaper than GPT-4o, long-context use no longer adds extra fees, and prompt cache discount rates are set at 75%.
Due to having GPT-4.1 as the standard now, the GPT-4.5 Preview will be discontinued by OpenAI on July 14, 2025.