OpenAI launches GPT-4.1 models with major performance gains

On Monday, OpenAI launched its latest AI model, GPT-4.1, along with smaller versions, GPT-4.1 mini and GPT-4.1 nano, lauding significant advances in coding, instruction following, and long context comprehension.

Outperforms GPT-4o and GPT-4.5 in Coding, Comprehension, and Cost

OpenAI has unveiled a new family of models, known as GPT-4.1, which are now available exclusively through the company’s application programming interface (API). According to the company, these models outperform the highly capable GPT-4o across the board.

Improved Capabilities with Lower Costs

The GPT-4.1 models bring significant advancements in context understanding, capable of handling up to 1 million tokens — a major leap in long-context comprehension. The models also come with updated knowledge through June 2024, making them more relevant for current use cases.

Crucially, OpenAI stated that these models operate at a “much lower cost” than their GPT-4.5 predecessors. As a result, the company plans to deprecate GPT-4.5 preview access in the API by July, citing that GPT-4.1 offers “improved or similar performance.”

Also read: Lady Gaga & Bruno Mars’ “Die With a Smile” hits 16 Weeks at No. 1 on Billboard Global 200

Massive Performance Gains for Developers and AI Agents

When it comes to specific benchmarks:

• GPT-4.1 showed a 21% improvement over GPT-4o
• A 27% improvement over GPT-4.5 in coding tasks
• Marked enhancements in instruction following and long-context understanding

These improvements make GPT-4.1 particularly well-suited for AI agents and advanced automation.

Real-World Utility at the Forefront

Despite strong benchmark results, OpenAI CEO Sam Altman emphasized practical benefits over numbers.

“Benchmarks are strong, but we focused on real-world utility, and developers seem very happy,” he said in a post on X.

Background: Transition from GPT-4.5 Preview

Earlier this year, OpenAI introduced the GPT-4.5 research preview to select users. The rollout of GPT-4.1 signals a shift toward more powerful and cost-efficient models for production-level applications.