OpenAI LP today made GPT-4, its newest and most capable language model, generally available through a cloud-based application programming interface.

The heavily funded artificial intelligence startup also detailed several other product updates. Alongside GPT-4, OpenAI is making three other AI models’ APIs generally available. Additionally, the startup announced plans to phase out a number of earlier neural networks.

OpenAI debuted GPT-4 in March as the successor to GPT-3, the large model that catapulted the startup into industry prominence three years prior. GPT-4 interprets user instructions more effectively, can take images and as input and features enhanced reasoning capabilities. The latter upgrade allows the model to perform complex tasks such as solving math problems more easily.

The GPT-4 API that became generally available today allows developers to integrate the model into their applications. According to OpenAI, the service is now accessible to all customers who have previously purchased access to its APIs. 

The startup will open access to new developers later this month. Further down the road, OpenAI intends to increase the API’s usage limits. The company’s long-term plans, meanwhile, include the launch of a so-called fine-tuning capability, which will enable developers to enhance GPT-4’s accuracy by training it on custom datasets.

GPT-4 launched into general availity today alongside the APIs of three other OpenAI models.

The first is GPT-3.5 Turbo, a predecessor to GPT-4 that offers more limited capabilities for a significantly lower cost. OpenAI is also opening access to the APIs of its DALL-E and Whisper models. The latter two neural networks are optimized for image generation and speech transcription, respectively.

On the occasion, OpenAI also disclosed plans to discontinue a number of earlier language models that it sells via APIs. They are set to go offline on Jan. 4, 2024. Some of the affected models will be automatically upgraded to newer OpenAI products, while others will require developers to manually make the switch.

The changes relate in part to an update that the startup rolled out to its API technology earlier this year.

Many of the models that are set to be phased out use a programming interface OpenAI calls the Completions API. It originally became available in 2020. This past March, OpenAI debuted a newer programming interface called Chat Completions API that will become the primary focus of its development efforts going forward.

“The initial Completions API was introduced in June 2020 to provide a freeform text prompt for interacting with our language models,” OpenAI detailed in a blog post today. “We’ve since learned that we can often provide better results with a more structured prompt interface.”

The new Chat Completions API introduces several improvements. Most notably, it provides mitigations against prompt injection attacks, which are attempts by hackers to modify an AI model’s output using malicious instructions. The enhancement should help developers improve the security of their OpenAI-powered machine learning applications. 

 Image: Unsplash