- Technology
- AI and automation
September 3, 2020updated 04 Sep 2020 3:45pm
If you want to sign up you might need deep pockets...
By Matthew Gooding
Updated 04/09/20: Comment added from OpenAI
The highly public beta testing (including, yes, something akin to the Turing Test) has been done: now OpenAI has revealed the preliminary pricing structure for users of its hugely powerful AI system — and it does not look cheap.
From October 1, users of the OpenAI API, which connects users to the organisation’s natural language processing (NLP) AI tools, will be able to choose from four different pricing plans, users signed up to the beta said this week.
The API, OpenAI’s first commercial product, is the gateway to an AI of unprecedented ability. Trained on the organisation’s GPT-3 autoprogressive NLP model, it has processed an enormous volume of data from books, academic papers and the internet. With 175 billion parameters it can power advanced chatbots, write news stories based on a small snippet of text, and even automate parts of the litigation process.
According to reports, OpenAI announced the pricing scheme for GPT-3’s API usage from October.
View all newsletters Sign up to our newsletters Data, insights and analysis delivered to you By The Tech Monitor team h/t @gwern @sonnylazuardi pic.twitter.com/w7GlkyDbLV
Content from our partners
DTX Manchester welcomes leading tech talent from across the region and beyond
The hidden complexities of deploying AI in your business
When it comes to AI, remember not every problem is a nail
— hardmaru (@hardmaru) September 3, 2020
Tokens, Tokens Everywhere
Free access to the API will be limited to 100,000 “tokens” (or a three month trial, whatever comes first) from October 1, according to a message sent to beta API users.Users can purchase two million tokens for $100 per month, or ten million for $400 a month, and (of course) there’s also a bespoke tier for anyone who wants to take their usage to the next level.
See also:Bank of America Chatbot Tops 10 Million Users
What does this mean in practice?
Well, OpenAI says two million tokens will be sufficient to process 3,000 pages of text. For context, it adds that you could process the entire works for Shakespeare using 1.2 million tokens.
That’s downright reasonable. I can’t wait to play with it.
— Fraser Cain (@fcain) September 3, 2020
Murat Ayfer is the developer of PhilosopherAI, which uses the OpenAI API to generate interesting texts based on a single keyword or phrase.
PhilosopherAI uses the OpenAI API to generate interesting text
He told Computer Business Review: “So far PhilosopherAI.com has had 850,000 queries entered and generated 612,949,783 characters.
“This would cost around $6,000 if it was all generated in one go, but the background prompt juggling I do probably multiplies it by three, putting me at close to $18,000 already.”
Murat currently offers PhilosopherAI free of charge, but is looking at the viability of different pricing models.
He added: “The only way the pricing will make sense is that the cost of the API is offloaded to the user. This means only things that consumers really demand will survive in the short term. I think it will put good pressure on developers working with GPT and will provide a lot of clarity on what it *really* is good for.
“It’s going to be interesting now reality is setting in. A lot of the hyped up use cases are currently quite costly; would you pay five or ten cents for the summary of an article? Some will, some won’t.”
OpenAI has been approached for comment on the pricing structure. The non-profit organisation, backed by Microsoft, had to pull the plug on an earlier free version of the system amid fears it would be used to generate fake news and other malicious applications.
OpenAI told Computer Business Review: “OpenAI is starting to demo pricing for current beta users of the API, which goes into effect October 1. This pricing includes four tiers: Explore, Create, Build, and Scale. We will continue to evaluate pricing during the beta period.
“The API remains in a limited private beta, and we have no schedule for GA at this time as we continue to carefully evaluate how our models perform in the real world.”
Read More: Hackers Offer Tesla Staffer $1M to Plant Malware
Topics in this article : AI , Homepage News List , Homepage Top Feature , NLP , OpenAI
Matthew Gooding
Matthew Gooding is the former managing editor of Tech Monitor. A former business editor of the Cambridge News, he has particular interests in the semiconductor industry and cloud computing.
