ChatGPT officially revealed the paid version, not costing 1 million VND/month as rumored but surprisingly cheaper

After a while of hiding, OpenAI has recently officially announced the premium version of ChatGPT Plus, allowing users to have priority to use the AI ​​chatbot that is creating a fever around the world thanks to its ability to reply quickly. flows and very naturally all questions in many areas that users ask.

Accordingly, ChatGPT Plus will operate on a monthly fee model. When purchasing and using ChatGPT Plus, users will still use the service normally even during peak times, while free users will have to wait a while for their turn. In addition, ChatGPT Plus will also provide users with “faster response times” and “priority access to new features and improvements”.

However, OpenAI has yet to reveal whether ChatGPT Plus is ‘smart’ and gives more accurate answers than the free version. Interestingly enough, the paid version of ChatGPT used to be ” trained on a much larger data set and fine-tuned to perform specific tasks like dialogue generation, summarization, and language translation.” It can also generate more human-like responses, making it useful for applications like customer service, personal assistants, and chatbots ,” according to the response given by the chatbot itself.

Reportedly, OpenAI said it will send invitations to use the service to people in the US and on the company’s waiting list “in the coming weeks”. At the same time, OpenAI is expected to expand the implementation of the toll model to other countries and regions in the future.

A very remarkable point is that the user cost to use ChatGPT Plus is only about 20 USD/month (ie 469 thousand VND) . This price announced by OpenAI is much lower than the previously rumored figure of 42 USD (about 1 million VND).

According to The Verge, 42 USD is a price that can make users feel hesitant, especially users who just want to experience and do not plan to take advantage of ChatGPT to make money.

Of course, at a price of 20 USD, OpenAI can still collect revenue to maintain the service, even be profitable, while still maintaining the attractiveness of ChatGPT to users. OpenAI suggests that ChatGPT Plus could be one of many paid packages that will be revealed by the company in the near future. In the blog post, OpenAI claims it is “actively exploring” options for lower-paid plans, customized to each user’s needs.

It must also be added that at the present time, the $20 price tag itself will also set a standard price threshold for the upcoming paid AI chatbots, which are almost certainly about to hit the market. With OpenAI almost at the forefront of bringing AI to the masses, any party trying to launch a chat bot that costs more than $20/month will have to explain why their AI is worth it. more expensive than ChatGPT Plus.

Of course, ChatGPT will still maintain the free version. OpenAI said it will continue to provide free access to ChatGPT, insisting paid users will “help support free accessibility for as many people as possible”.

Previously, as admitted by Sam Altman, CEO of OpenAI, processing all queries from a number of users cost the company a lot of computing costs. This is understandable, as ChatGPT needs a large amount of computing resources to handle user inquiries.

Accordingly, the cost of running OpenAI’s ChatGPT is said to be about 100,000 USD per day. According to many sources, Microsoft’s Azure cloud computing service has integrated ChatGPT so that OpenAI does not have to invest in physical servers.

Considering Microsoft’s current pricing, OpenAI costs about $3 an hour for an A100-specific GPU, with each word generated on ChatGPT costing $0.0003.

At least eight A100 GPUs are being used to work on a single ChatGPT. So, when ChatGPT generates an average response of 30 words, OpenAI will have to pay close to 1 cent. Through such an estimate, OpenAI can spend at least 100,000 USD per day or 3 million USD monthly on operating costs.

Another source claims that a single NVIDIA A100 GPU can run a 3 billion parameter model in about 6ms. Taking this speed as a reference point, an NVIDIA A100 GPU can take 350 milliseconds to generate just one word on ChatGPT.

Since the latest version, ChatGPT-3.5, has more than 175 billion parameters, it will take about 5 GPU A100 computations to get the output for one query. The average number of at least 8 A100 GPUs is calculated later based on ChatGPT’s ability to generate about 15-20 words per second. It can be clearly seen that the cost is so great that OpenAI is forced to collect fees from ChatGPT.