One of the leading artificial intelligence (AI) companies, OpenAI, has introduced upgraded AI models – GPT-4.1. This once again proves the progress of AI and the continuous steps forward in improving and enhancing the multimodal processing capabilities of such products, opening even wider and higher-quality AI applications across various industries.

OpenAI’s New GPT-4.1 Models Significantly Improved

On April 14, OpenAI unveiled a new series of GPT models with key improvements in coding, instruction-following, and long-context capabilities. Alongside this, the company introduced another innovation – its first nano model.

In total, the new series presents 3 new API models: GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano. According to the company, these models outperform GPT-4o and GPT-4o mini in performance, enabling improvements in code-related tasks.

In general, the models also feature larger context windows that support up to 1 million tokens, allowing them to better understand large-scale context. This essentially equals processing 750,000 words at once.

Along with the new model, smaller and more affordable versions were also introduced, which are claimed to be the smallest, fastest, and most affordable model variants.

Improvements Supported by Numbers

According to OpenAI, new model demonstrates strong performance in standard benchmarks:

1. Coding. GPT-4.1 achieves a 54.6% score in the SWE-bench Verified benchmark, which is 21.4% higher than GPT-4o, and it also outperforms GPT-4.5. These results position the new OpenAI model as one of the best in the field of programming.
2. Instruction following. Based on the Scale MultiChallenge benchmark, which evaluates how well a model understands and follows instructions, the new model scores more than 3 times higher than GPT-4o. This indicates that the new version promises more accurate and faster response generation, along with the ability to process multimodal content, text, audio, or images in real time.
3. Long-context understanding. The new OpenAI model also shows impressive results in multimodal long-context comprehension tasks. According to the Video-MME benchmark, which tests models’ understanding of long video contexts, the new model outperforms the current one by nearly 7%.

Capabilities of GPT-4.1 mini and GPT-4.1 nano Models

GPT-4.1 mini shows significant performance progress in the small AI model category. It fully matches or exceeds GPT-4o in intelligence evaluations, while cutting latency nearly half and costs by 83%.

GPT-4.1 nano is the fastest and most affordable model offered by OpenAI for tasks that require low latency, such as classification or autocompletion.

GPT-4.1 Capabilities Already Tested in Real Use Cases

For example, in the windsurfing field, the new model was tested in internal coding processes and achieved 60% higher efficiency than GPT-4o.

The instruction-following feature was tested by Blue J, where the new model outperformed the current one by more than half in accuracy, based on Blue J’s internal benchmark of the most complex real-world tax scenarios.

Meanwhile, the long-context capabilities were tested by Carlyle, who attempted to extract multiple granular financial data points from lengthy document files such as PDFs, Excel files, and others. Compared to other models, the company reported that the new model performed 50% better in retrieving data from very large, data-dense documents.

Focus on Developers

The new models are available only via the company’s API, not directly through ChatGPT. This means the new model is aimed at developers, who are increasingly competing to build efficient models capable of performing complex programming tasks. Public discourse also includes opinions suggesting that this OpenAI model could eventually replace humans in certain coding processes.

These new improvements are expected to be gradually integrated into the latest GPT-4o version.

Final Word

By releasing the new GPT-4.1 AI model series, OpenAI opens up even broader possibilities for deep and efficient AI model operation in handling multimodal data with the highest possible accuracy and quality in real time.

At the same time, the smaller models make AI solutions more accessible to small businesses or those with limited financial resources. This benefits not only business participants but also end-users who directly experience the results of such innovative solution integrations.

If you are interested in this topic, we suggest you check our articles:

• OpenAI Strives for the Most Human-Like Voice
• Sam Altman: OpenAI CEO Background and Rise to the Top
• Can xAI Grok 3 rival OpenAI?
• What Are the New OpenAI Tools Launched to Help AI Agent Development?
• OpenAI and Anthropic Compete for College Students with Free AI Tools

Sources: CoinGeek, OpenAI, TechCrunch