OpenAI is Lowering the Costs of its Models

OpenAI has announced several new updates, including a series of new embedding models and that it is reducing prices for GPT-3.5 Turbo for the third time in a year.

GPT-3.5 Turbo input prices have now been reduced by 50% to $0.0005 per 1k tokens and output prices are reduced by 25% to $0.0015 /1K tokens.

That would put GPT-3.5 Turbo at a cost per 1k tokens less than Anthropic’s Claude 2.0 and 2.1, which costs $0.008/1K token inputs and $0.024/1K token outputs.

OpenAI also announced it’s introducing an updated version of GPT-3.5 Turbo, gpt-3.5-turbo-0125, which boasts improvements including higher accuracy at responding in requested formats and a fix for a bug that caused a text encoding issue for non-English language function calls.

The company is also releasing the updated GPT-4 Turbo model, gpt-4-0125-preview. The preview model can complete tasks more thoroughly than prior models and boasts a larger context window, around 85,333 words per input.

The new GPT-4 Turbo preview model also has lower prices.

A GPT-4 Turbo with vision capabilities will be coming to general availability “in the coming months,” OpenAI also announced.

New embedding models

OpenAI also unveiled new embedding models. An embedding model is a type of AI model that transforms high-dimensional data, like text or images, into a lower-dimensional space, allowing for easier computation of similarities between data points.

Related:ChatGPT Sued for Libel: Is There Freedom of Speech for AI?

OpenAI revealed text-embedding-3-small, a new embedding model meant as an upgrade to its predecessor, text-embedding-ada-002, which was released in December 2022.

The new embedding model has improved abilities at multi-language retrieval and English language tasks.

It’s also five times cheaper to use than text-embedding-ada-002, available at a price per 1k tokens of just $0.00002.

There’s also a new large-scale embedding model: text-embedding-3-large. Boasting a stronger performance, the larger embedding model is priced at $0.00013 per 1k tokens.

Both of OpenAI’s new embedding models were trained using a new technique that allows developers to trade off performance and cost of using embeddings. Developers can now shorten embeddings without the embedding losing its concept-representing properties.

New moderation model

OpenAI also unveiled text-moderation-007, a new moderation model for identifying potentially harmful text.

The ChatGPT maker is trying to update its safety efforts in the wake of its boardroom shakeup. The new moderation model enables developers to check their content against OpenAI's usage policies to detect potentially problematic outputs.

Related:From Davos: Sam Altman on AGI, NYT Lawsuit and Getting Fired