OpenAI Launches GPT-4.1, Offering Upgrades in Coding, Context Processing, Efficiency
- By John K. Waters
- 04/23/25
OpenAI has introduced GPT-4.1, offering stronger performance across software development, instruction following, and long-context comprehension.
The newly introduced lineup — GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano — expands OpenAI's API offerings with a focus on cost-effectiveness, lower latency, and greater intelligence. These models were designed to empower autonomous agents and scalable applications that can perform complex tasks across domains, such as legal analysis, customer support, and code generation.
"We trained these models with a focus on real-world utility," OpenAI said in a blog post.
Understanding the Evolution: GPT-4.5, GPT-4o, and GPT-4.1
As OpenAI's model lineup has expanded, so has the complexity of its naming and release strategy. The shift away from a strict version-numbering scheme and the use of codenames like "Orion" have contributed to blurred lines between model generations. GPT-4.5 may have been an internal stepping stone, while GPT-4.1 integrates and formalizes many of its capabilities. The takeaway: GPT-4.1 is not just a point release — it is the production-ready culmination of several iterative advances.
Smarter Code, Faster Output
At the core of GPT-4.1's appeal is its exceptional performance in software engineering. It achieves 54.6% accuracy on SWE-bench Verified — up from 33% for GPT-4o — while also excelling in multi-language code editing tasks.
Windsurf, an early tester, reported 30% better efficiency in tool use and a 50% reduction in redundant edits, speeding up development cycles considerably.
Better at Following Instructions
GPT-4.1 also improves instruction compliance, particularly for multi-turn and format-sensitive prompts. It scores 38% on Scale AI's MultiChallenge and outperforms earlier models in OpenAI's internal instruction-following evaluations.
Legal tech firm Blue J saw a 53% jump in complex scenario comprehension, while Hex reported nearly double the accuracy in executing SQL queries and handling ambiguous schemas.
Long Context, Low Friction
All GPT-4.1 models support up to 1 million tokens of context, enabling AI to analyze, reference, and respond based on extensive inputs — such as full legal contracts or massive code repositories.
Benchmarks like OpenAI-MRCR and Graphwalks confirm GPT-4.1's superiority in retrieving nuanced information from long inputs and performing multi-hop reasoning. Thomson Reuters reported a 17% improvement in legal clause cross-referencing, and Carlyle saw a 50% accuracy boost in extracting data from large financial reports.
Speed and Cost for Every Use Case
GPT-4.1 mini cuts latency in half while maintaining intelligence comparable to GPT-4o.
GPT-4.1 nano, ideal for mobile and lightweight inference, delivers responses in under five seconds and offers the best price-performance ratio to date.
These improvements are further supported by pricing updates: GPT-4.1 is 26% more cost-efficient than GPT-4o on average, and long-context usage no longer carries additional fees. Prompt caching discounts have increased from 50% to 75%.
Transitioning from GPT-4.5
As GPT-4.1 becomes the new standard, OpenAI is retiring GPT-4.5 Preview on July 14, 2025. The company noted that while GPT-4.5 helped explore ambitious capabilities, GPT-4.1 brings those into full production maturity.
For more information, read the OpenAI blog.
About the Author
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at [email protected].