OpenAI has released ‘ChatGPT’ API (Application Programming Interface).

Because of this, general firms and developers can even develop applications that run ChatGPT by utilizing OpenAI’s language and voice-to-text conversion functions.

TechCrunch reported on the first (local time) that OpenAI has released a latest ChatGPT API based on the ‘GPT-3.5-Turbo’ model, which has improved speed by optimizing ‘GPT-3.5’.

The brand new ChatGPT API costs $0.002 per 1000 tokens or 750 words, 10x cheaper than GPT-3.5. OpenAI plans to proceed improving the GPT-3.5-Turbo model.

For developers who anticipate using more tokens than the usual API allows, we now have also launched the ‘Foundry’, an enterprise developer platform to pay for allocating dedicated compute infrastructure to run the GPT-3.5-Turbo model. did.

Dedicated compute infrastructure can use the GPT-3.5-Turbo model with as much as 16k context windows. This enables for about 450 million tokens per day, 4 times greater than the usual ChatGPT API.

OpenAI said firms similar to Instacart, Shopify, Quizlet and Snap use the ChatGPT API. Snap can be the primary customer of Foundry, an enterprise developer platform.

Meanwhile, in September of last 12 months, OpenAI optimized Whisper, an open-source speech-to-text model, and released a hosted version, Whisper API. For $0.006 per minute, it will probably be used to transcribe or translate audio from multiple languages ​​into English.

Technically, the Whisper model is open source, so you may run it on your individual hardware without paying anything. Nevertheless, in case you want fast processing or you must perform transcription on a low-powered device like a phone, using the OpenAI API is effective.

Chan Park, cpark@aitimes.com