OpenAI Debuts Cheaper GPT-4.1 Models to Attract Enterprise Clients

OpenAI has launched its newest artificial intelligence (AI) model lineup that is less expensive than older models and also outperforms them, as competition for enterprise customers heats up.

The new models — GPT-4.1, GPT-4.1 mini and GPT-4.1 nano — also boast one-million-token context windows, up from 128,000. Context windows are where users enter prompts and get responses. One million tokens equal around 750,000 words, or about seven to 10 novels.

In a blog post Monday (April 14), OpenAI said the GPT-4.1 models outperform its own previous reasoning models GPT-4o and GPT-4o mini “across the board,” with “major gains” in coding capabilities, how well it follows instructions and ability to handle long documents. It was trained on knowledge up to June 2024.

Like the GPT-4o series, these new models can handle text, images, audio and video. They are available now, but only as an application programming interface (API) for developers.

“One of the things that we’ve learned over and over is that the more cost effectively we can offer our models, the more use cases you’re able to build, the more that you’re able to use AI to help people all over the world,” said OpenAI Chief Product Officer Kevin Weil in a YouTube video.

The cost of using AI is a major stumbling block to AI deployment at enterprises. Each prompt and response using an AI chatbot or assistant carries a cost per token. That’s why Chinese AI startup DeepSeek’s inexpensive open-source foundation AI models went viral.

Notably, the $20 monthly ChatGPT subscription for unlimited use is a loss leader to get people used to using AI chatbots. OpenAI CEO Sam Altman said on X that even the $200 a month ChatGPT Pro plan loses money. For enterprises, OpenAI has a pay-as-you-go plan.

AI is expensive because of the computer chips it uses, according to Amazon CEO Andy Jassy’s 2024 letter to shareholders. “Most AI to date has been built on one chip provider,” referring to Nvidia, he wrote. “It’s pricey.”

But Jassy sees costs dropping as other companies build AI chips. He said Amazon’s own Trainium2 chips are 30% to 40% cheaper to use.

Read more: OpenAI to Release GPT-4.5 Within Weeks, GPT-5 Within Months

OpenAI said GPT-4.1 is 26% less expensive than GPT-4o for median queries, while GPT-4.1 nano is its cheapest and fastest model to date.

OpenAI also raised the prompt caching discount to 75% from 50%. That means when users upload a document and ask several questions related to it, OpenAI will charge 75% less for the second and following prompts. There is an additional 50% discount for Batch API uses as well.

OpenAI shared the following real-world use cases of GPT-4.1:

• Thomson Reuters used GPT-4.1 with its legal AI assistant, CoCounsel, and found a 17% improvement in accuracy when reviewing multiple, lengthy documents.
• Investment firm Carlyle reported a 50% improvement on retrieval from large documents with dense data from PDFs, Excel files and other complex formats.
• Software company Windsurf said GPT-4.1 scores 60% higher than GPT-4o on their internal coding benchmarks, with users noting 30% more efficient tool calling and about 50% fewer unnecessary edits.
• Tax research firm Blue J said GPT-4.1 was 53% more accurate than GPT-4o on its most challenging tax scenarios.
• Data analytics company Hex saw a nearly two-fold improvement on its most challenging SQL evaluation set.

The post OpenAI Debuts Cheaper GPT-4.1 Models to Attract Enterprise Clients appeared first on PYMNTS.com.