OpenAI Launches GPT-4.1 Models with Advanced Coding Capabilities
OpenAI has unveiled a new family of AI models specifically optimized for coding, strengthening its position against competitors like Google and Anthropic. The three new models—GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano—are now available to developers through OpenAI's API.
NEWS
4/24/20251 min read
According to Chief Product Officer Kevin Weil, these models outperform OpenAI's widely-used GPT-4o and, in certain aspects, even surpass their most powerful model, GPT-4.5. GPT-4.1 achieved a 55% score on SWE-Bench, a standard coding benchmark, placing it several percentage points above OpenAI's previous models.
"These models excel at coding, complex instruction following, and building agents," Weil stated during the announcement livestream.
Key improvements include:
Ability to analyze 8x more code simultaneously, enhancing bug-fixing capabilities
40% faster processing than GPT-4o
80% reduction in input query costs
Better instruction comprehension, minimizing repetitive prompting
Windsurf CEO Varun Mohan, whose company tested GPT-4.1, reported it performed "60% better" than GPT-4o on their internal benchmarks, with "substantially fewer cases of degenerate behavior" and less time wasted on irrelevant files.
The release comes amid intensifying competition in the AI landscape. Stanford University's recent report on AI indicates that models from Google and DeepSeek now rival OpenAI's capabilities, with Anthropic, Meta, and Mistral following closely.
Despite reporting 500 million weekly active users with rapid growth, OpenAI faces pressure to build a sustainable business model while continuing to invest heavily in research and infrastructure. The company recently announced plans to release an open-weight model this summer, allowing developers to download and modify it for free.