OpenAI has launched three new models via API: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. These models outperform GPT-4o and GPT-4o mini in coding, instruction following, and long-context comprehension. Each supports a 1 million-token context window and features an updated knowledge cutoff of June 2024.

Stronger Benchmark Results in Coding and Instruction Following

GPT-4.1 scores 54.6% on SWE-bench Verified, surpassing GPT-4o by 21.4 percentage points and GPT-4.5 by 26.6 points. In Scale’s MultiChallenge benchmark, it reaches 38.3%, a 10.5-point improvement over GPT-4o. On the Video-MME long-context test without subtitles, GPT-4.1 sets a new high at 72.0%.

Enhanced Model Utility with Lower Cost and Latency

GPT-4.1 mini improves performance among small models, often exceeding GPT-4o in benchmarks. It reduces latency by nearly 50% and costs 83% less while matching GPT-4o in intelligence evaluations. GPT-4.1 Nano, the fastest and most affordable option, is ideal for low-latency tasks like classification and autocompletion.

Improved Reliability for Building AI Agents

Instruction following and long-context comprehension make GPT-4.1 more capable for AI agents. Paired with tools like the Responses API, developers can build systems to handle software engineering, document analysis, and customer support with minimal guidance.

GPT-4.5 to Be Deprecated in Favor of GPT-4.1

GPT-4.5 Preview will be discontinued on July 14, 2025. GPT-4.1 offers similar or improved performance at lower cost and latency. OpenAI notes that GPT-4.5 served as a research release, and its valued features will be carried into future models.

Visual Understanding and Cost Efficiency Improvements

GPT-4.1 mini outperforms GPT-4o in image benchmarks and excels in multimodal tasks. In the long-format Video-MME test without subtitles, GPT-4.1 scores 72.0%, up from GPT-4o’s 65.3%. GPT-4.1 is also 26% cheaper than GPT-4o for average queries. Prompt caching discounts rise from 50% to 75%, and long-context usage has no added cost beyond token pricing.

PHOTO: GETTY IMAGES

This article was created with AI assistance.