A juggernaut has emerged in artificial intelligence. OpenAI’s GPT-4.1 isn’t just another incremental update—it’s crushing benchmarks left and right. The successor to GPT-4o and experimental GPT-4.5 comes in three flavors: GPT-4.1, mini, and nano. Each packs serious punch with knowledge updated through June 2024.

Let’s talk coding. GPT-4.1 scores 54.6% on SWE-bench Verified, demolishing GPT-4o’s measly 33.2%. It navigates repos better, passes more tests, and human reviewers preferred its web apps 80% of the time. Not bad.

The model follows instructions like a well-trained puppy. Scoring 87.4% on IFEval (compared to GPT-4o’s 81.0%), it handles complex directives, negative instructions, and ordered steps with surprising competence. Its 90.2% on MMLU puts it near the top of the leaderboard, though OpenAI’s newer o1-preview achieved an even higher MMLU score of 90.8.

Perhaps most impressive? Context length. All GPT-4.1 variants now support a million tokens. That’s enough to analyze entire codebases or novels in one go. Like the original GPT-4o, this model excels with multimodal capabilities that process text, images, and audio inputs simultaneously. It retrieves information from massive documents with remarkable accuracy, setting records on Video-MME and MRCR benchmarks.

Speed and cost improvements matter too. GPT-4.1 runs 26% cheaper than GPT-4o for typical queries. The mini version slashes latency nearly in half while being 83% cheaper—yet still matches or beats GPT-4o on most tests. Crazy. The nano variant is blazing fast for simpler tasks like classification.

OpenAI even boosted prompt caching discounts to 75%, up from 50%. Smart move.

How does it stack up against competitors? Pretty damn well. It outperforms previous OpenAI models across the board. While Claude and Gemini might edge it out slightly on specific tests, GPT-4.1’s balanced performance across coding, reasoning, and long-context tasks makes it a formidable contender.

For developers already using GPT-4o, the upgrade is a no-brainer. Better performance at lower cost? Yes, please. The AI arms race continues, and OpenAI just fired a serious shot.