OpenAI turbocharges GPT-4 and makes it cheaper

OpenAI introduced extra enhancements to its massive language fashions, GPT-4 and GPT-3.5, together with up to date data bases and a for much longer context window. The corporate says it can additionally comply with Google and Microsoft’s lead and start defending prospects in opposition to copyright lawsuits.

GPT-4 Turbo, presently obtainable through an API preview, has been skilled with data relationship to April 2023, the corporate introduced Monday at its first-ever developer conference. The sooner model of GPT-4 released in March solely realized from information dated as much as September 2021. OpenAI plans to launch a production-ready Turbo mannequin within the subsequent few weeks however didn’t give an actual date.

GPT-4 Turbo will even “see” extra information, with a 128K context window, which OpenAI says is “equal to greater than 300 pages of textual content in a single immediate.” Typically, bigger context home windows let massive language fashions like GPT perceive extra of the query and supply extra thought-out responses. Beforehand, OpenAI launched two variations of GPT-4, one with a context window of solely 8K and one other at 32K. 

OpenAI says GPT-4 Turbo is cheaper to run for builders. Enter will price solely $0.01 per 1,000 tokens — the essential unit of textual content or code for LLMs to learn — in comparison with the $0.03 on GPT-4. Every output will probably be $0.03 per 1,000 tokens. Total, OpenAI says the brand new model of GPT-4 is 3 times cheaper than the sooner ones.

This latest model of GPT-4 will nonetheless settle for picture prompts, text-to-speech requests, and combine DALL-E 3, a function first announced in October. 

The corporate says the enhancements to GPT-4 Turbo imply customers can ask the mannequin to carry out extra complicated duties in a single immediate. Folks may even inform GPT-4 Turbo to particularly use the coding language of their selection for outcomes, like code in XML or JSON. 

Firms utilizing the GPT-3.5 Turbo mannequin will even see enhancements in context home windows, performance, and pricing. It should have a 16K context window by default and the identical operate updates as GPT-4 Turbo. GPT-3.5 Turbo will price $0.01 for enter and $0.002 for output. 

Past enhancements to the flagship mannequin, OpenAI additionally introduced it can comply with within the footsteps of Microsoft and Google and supply copyright indemnity to enterprise customers by way of a program referred to as Copyright Protect. 

“We’ll now step in and defend our prospects, and pay the prices incurred, should you face authorized claims round copyright infringement,” the corporate stated in an announcement. 

Copyright Protect will cowl usually obtainable options of ChatGPT Enterprise and OpenAI’s developer platform.