OpenAI Unveils GPT-4.1: 8x More Context and Lower Costs

OpenAI has introduced three versions of its flagship AI model: GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano. All of them support processing up to 1 million tokens of context, spanning text, images, and video. That’s eight times more than GPT-4o, which handled up to 128,000 tokens. According to OpenAI, GPT-4.1 is better at handling long prompts, extracting key information more accurately, and filtering out irrelevant details.

The update places special emphasis on improvements in code generation and instruction following. For example, GPT-4.1 successfully solves 54.6% of tasks in the SWE-Bench benchmark, showcasing solid gains in programming capabilities. API usage is also now 26% cheaper than with GPT-4o — a direct response to rising competition from budget-friendly alternatives like DeepSeek.

GPT-4.1 and its lightweight variants are available exclusively via API. They won’t be showing up in the public ChatGPT app. Meanwhile, OpenAI has started sunsetting older models:

According to the company, GPT-4.1 “matches or outperforms its predecessors on key metrics while being more cost-efficient”. The announcement comes as the release of GPT-5 gets pushed back. Initially expected in May, GPT-5 is now delayed, with Sam Altman citing integration challenges. At the same time, new logic-focused models, o3 and o4, are expected soon — early references have already been spotted in ChatGPT’s code.