OpenAI has introduced a new line of GPT-4.1 artificial intelligence models, including GPT-4.1, GPT-4.1 mini, and their first-ever nanomodel, GPT-4.1 nano. All three AI models are available via the OpenAI API and are not integrated into ChatGPT.

About new GPT-4.1 AI models

  • These are multimodal models with an extended context of 1 million tokens (about 750,000 words) that the company says are excellent at programming and executing instructions.
  • GPT-4.1 models are a response to competition from Google and Anthropic, which have also recently introduced powerful models for programming. In particular, Google has released Gemini 2.5 Pro with a similar context window of 1 million tokens, which ranks high in popular coding benchmarks, and Anthropic has released Claude 3.7 Sonnet. Other competitors include Chinese startup DeepSeek with an updated V3 model.
  • OpenAI positions GPT-4.1 as an improved version of GPT-4o and GPT-4o min. According to the company's internal tests, GPT-4.1 shows better results in coding benchmarks, including SWE-bench, and the company positions it as an improved version of GPT-4o and GPT-4o min.

OpenAI GPT-4.1 model specsImage: OpenAI

  • The mini and nano models are fast and economical, but inferior in accuracy, and OpenAI claims the GPT-4.1 nano is its fastest and cheapest model ever.

New models prices:

  • GPT-4.1 - $2 per million incoming tokens and $8 per million outgoing tokens;
  • GPT-4.1 mini - $0.40 / $1.60 respectively;
  • GPT-4.1 nano - $0.10 / $0.40.

Despite its successes, OpenAI recognizes that GPT-4.1 has limitations: the model becomes less accurate as the amount of input data increases. For example, in the OpenAI MRCR test, accuracy drops from 84% with 8,000 tokens to 50% with 1 million tokens.

OpenAI is phasing out GPT-4.5 for developers

The company has also announced that it will stop supporting its largest GPT-4.5 model in the API as of July 14, 2025. This model, codenamed Orion, was released at the end of February. GPT-4.5 will remain available in ChatGPT in a trial mode for paid users.

GPT-4.5 was the most expensive of the OpenAI series: $75 per million inbound tokens and $150 per million outbound tokens. The high cost of operation was one of the reasons for refusing to support the model in the API.

Verification to access future AI models

According to a support page published on the company's website last week, OpenAI may soon require a verification process for organizations that want to access its most powerful models in the future.

Verification will require a government certificate issued by a country supported by OpenAI. A document can validate an organization every 90 days, and not all organizations will be eligible for validation, OpenAI says.

The company notes that this decision is related to security - to prevent misuse and leakage of intellectual property. 

Previously, OpenAI said it had evidence that Chinese AI startup DeepSeek was using its proprietary models to train a chatbot. 

As a reminder, OpenAI will soon replace the GPT-4 AI model in its popular ChatGPT chatbot. A new, more powerful GPT-4o model will be used instead.