GPT-3 is a large language model developed by the popular AI company OpenAI. This model is skilled at performing a variety of NLP tasks. This includes machine translation, question-answering, summarization, and more.
GPT-3 contains four language models Ada, Babbage, Curie, and Davinci that are capable of performing a variety of tasks. But these models are available at different prices per 1000 tokens.
In this article, we will take a look at How Much It Costs to Use GPT Models GPT-3 Pricing Explained. We will also talk about how you can measure token usage on GPT-3.
GPT-3 Pricing Explained
OpenAI has created a simple and flexible pricing setting for GPT-3. It contains four language models Ada, Babbage, Curie, and Davinci whose price is calculated based on per 1000 tokens.
So, Ada contains a price range of $0.0004 / 1K tokens. 100 tokens = 75 words, thus for $0.0004 users will be able to access 750 words using Ada. Similarly, you will be able to access the Davinci model for $0.0300 / 1K tokens.
Does GPT-3 cost money?
Yes, GPT-3 contains four models Ada, Babbage, Curie, and Davinci that contain different price ranges. The cost of GPT-3 starts from $0.0004 per 1000 tokens for the basic model Ada.
How much is GPT-3 vs GPT-4?
The cost of accessing GPT-3 and GPT-4 differs depending on what model you are choosing. GPT-3 contains four models Ada, Babbage, Curie, and Davinci that have a different price range per 1000 tokens.
Unlike GPT-3 the range of the GPT-4 model varies depending on the context you choose: Here is a table showcasing the price range of GPT-3 models:
| GPT-3 Model | Training Price | Usage Price |
| Ada | $0.0004 / 1K tokens | $0.0016 / 1K tokens |
| Babbage | $0.0006 / 1K tokens | $0.0024 / 1K tokens |
| Curie | $0.0030 / 1K tokens | $0.0120 / 1K tokens |
| Davinci | $0.0300 / 1K tokens | $0.1200 / 1K tokens |
Now that we have seen the price range of GPT-3 models, let’s take a look at GPT-4 models. Unlike GPT-3 the range of the GPT-4 model varies depending on the context you choose:
| GPT-4 Model | Input | Output |
| 8K context | $0.03 / 1K tokens | $0.06 / 1K tokens |
| 32K context | $0.06 / 1K tokens | $0.12 / 1K tokens |
How much does it cost to train the GPT-3 model?
It cost more than $4 million to train OpenAI’s large language model GPT-3 according to technologists and analysts.
Rowan Curran, a Forrester analyst focused on machine learning and Artificial Intelligence stated, more advanced large language models are capable of costing more than “the high-single-digit millions” to train.
Is GPT-3 free for personal use?
GPT-3 can be accessed free for personal use for three months using the free tokens. However, after three months, users need to purchase a token to continue using OpenAI’s GPT-3.
Is GPT-3 free forever?
No, GPT-3 is not available for free forever. OpenAI provides free tokens worth $5 to all new users, which can be accessed for the next three months. But once, users have utilized all free tokens or the three-month timeline is over. Users need to purchase further tokens to continue using GPT-3.
How much does it cost to use GPT-3 in a commercial project?
To ensure everyone can access GPT-3 comfortably, OpenAI has made its pricing flexible and simple. There are a total of four language models available: Ada, Babbage, Curie, and Davinci.
Among all four Davinci is considered to be the most powerful language model. This model is also used in the popular AI chatbot “ChatGPT.”
While the other three models Ada, Babbage, and Curie can be utilized for performing more simpler tasks like generating summaries or conducting sentiment analysis. The cost of the GPT-3 commercial project is calculated per every 1000 tokens.
These tokens can be assumed as pieces of words utilized for natural language processing. For generating an English text, 1 token will equal either four characters or 0.75 words. Davinci cost $1 for every 50K tokens used.
How to measure token usage in GPT-3?
To measure token usage in GPT-3, you need to combine prompts with text corpora. After combining the two, you need to send it to an API and calculate the total number of tokens returned.
The API request cost will then be monitored in the usage view. This will contain a one-request-per-one billing window limit, so you need to wait for around five minutes.
Then the cost will be calculated and further compared with the cost in the usage view. Below we have mentioned a step-by-step guide on how to measure the token usage in GPT-3.
Step 1: Estimating the price of GPT-3 inputs
To estimate the pricing of GPT-3 input, you need to first check the accuracy of the OpenAI pricing stated on the page. You can check the estimated pricing through “Tokenizer”, a tool that helps calculate how a piece of text will be tokenized by an API.
Along with the total token count in the piece of text. This way users can compare it with the data from the usage view and actual billing.
As the Corpora, you can take the top ten most downloaded apps and their descriptions. This includes apps such as Instagram, TikTok, Spotify, Facebook, Messenger, Telegram, Zoom, Snapchat, WhatsApp, and CpaCut.
This way you can perform a variety of operations on the following text and run through a corpora for several use cases. This includes keyword searching, summarizing, transforming texts, and more. The length of the text can be varied between 376 to 2060 words.
As our corpora, we took the descriptions of the ten most downloaded apps: TikTok, Instagram, Facebook, WhatsApp, Telegram, Snapchat, Zoom, Messenger, CapCut, and Spotify.
These would allow us to run several operations on the text and test the corpora for different use cases, such as keyword searching, summarizing longer pieces of text, and transforming the text into project requirements.
The length of the descriptions varied from 376 to 2060 words.
For example, if you entered a text sample with a description of the popular app TikTok. In the description, you can write what TikTok is, how it works, its features, and more.
Let’s assume the prompt consists of 1609 words and 2182 tokens. Then the GPT-3 model should cost around:
| Ada | Babbage | Curie | Davinci |
| $0,0009 | $0,0011 | $0,0044 | $0,0437 |
You can do the same with other apps to determine the cost of the GPT-3 model depending on the texts and tokens utilized.
Step 2: Preparing the prompts
Our next step is to create prompts. For this experiment, we will prepare three different prompts for three completely different use cases.
Prompt #1: Gathering project requirements with GPT-3
The first prompt is about gathering project requirements with GPT-3 based on the description of the given app.
“Express in detail, using points and bullet points, demands strictly related to the project of an application close to the below description.”
The prompt contains a total of 22 words and 148 characters which equals 26 Tokens. We have added all these values to the corpora and then calculated the estimated token usage again for each model.
Prompt #2: Writing a TL;DR summary with GPT-3
The next prompt is about creating a summary of long-form texts or paragraphs. For this, the model’s job is to recognize all the relevant parts of the texts and create a detailed concise recap. Here’s the prompt for the summary:
“Generate a brief summary of a paragraph containing all the takeaways of the text mentioned below.”
The above-mentioned prompt consists of 16 words and 99 characters. This equaled 18 tokens. The following values were added to corpora.
Prompt #3: Extracting keywords with GPT-3
In the third prompt, we will enter a prompt that can categorize the keywords from the text and make it available in a proper format.
“Divide the below content in search of keywords. Ensure the keywords are brief and concise. Allocate generic categories to each keyword, like person, date, place, number, year, day, year, and more. Showcase them in a list of categories: keyword pair.”
The following prompt consists of a total of 41 words and 250 characters. This equals 61 tokens.
It was 41 words (250 characters) long, which equaled 61 tokens. Together with the corpora text, it gave us:
In the next step, you are going to send out your prompts with corpora texts to the API. Calculate the number of tokens and then return in output and monitor the API requests in the usage view.
Step 3: GPT-3 API testing
For GPT-3 API testing we will look into the most popular language model of GPT-3 “Davinci.” In OpenAI the token usage can be measured on the five-minute timeframes.
You can try sending one API request every five minutes. Since every request will consist of a one-piece text which is a (corpora) and a prompt.
This way, users can get detailed information regarding the token usage for every combination. You can also later compare your results with the original estimates.
As an example, we will send around 30 combinations for the testing prompts with three prompts x ten app descriptions.
After sending all thirty requests. We will compare the results showcased on the Usage view with the ones that have been estimated from the metadata of the API calls.
The results will be coherent with each other. Apart from this, the token usage of the prompt will also turn out to be coherent with the usage estimated with Tokenizer.
You can also verify whether there is any correlation between the length of the input and the length of the output. To ensure that you can estimate the token usage of the output.
You will notice the correlation between the input and output tokens will turn out to be extremely weak. Thus, measuring the input tokens isn’t enough to estimate the total number of tokens used in a single request.
The slope varied between 0,0029 in the TL;DR summary. While the project requirement requests were 0,0246
What factors impact the cost of using GPT-3?
A variety of factors can impact the cost of using GPT-3 such as the quality of prompt, Customization level, Model temperature, and more. Here are the top factors that can affect the cost of utilizing the GPT-3 model:
Model’s temperature
A model’s temperature takes charge of the randomness of the generated output by the model. Setting the temperature at a higher value range can cost more unpredictable and diverse results.
This will impact your cost as well since higher computational resources will be needed to run the model at higher temperatures. Thus, you should edit your model’s temperature based on your costing and requirements.
Quality of prompt
This factor ensures the generated response is relevant and doesn’t generate wrong responses. So depending on the Quality of the prompt you select, there will be a variation in the price.
Availability
Another factor that can impact the cost of using GPT-3 is the availability of the model. Like, if the model you want to access is currently in high demand, then the cost might vary and will be priced higher depending on the limited availability.
Customization
Depending on your requirements for customization, the price of GPT-3 might be influenced. For example, if you want to access special functionality then further development might be required. The usage of additional development will result in increased costs.
However, you can control the budget by setting soft and hard limits on the platform. If you set a soft limit on your account, every time you pass a certain usage threshold, you will receive an email alert.
Meanwhile, if you set a hard limit, then your subsequent API request will be rejected once it reaches the usage threshold.
Summary
OpenAI has generated a simple and flexible price range for GPT-3, with the price range varying for each model. Users can access the popular model Davinci users need to purchase 1K tokens of $0.0300.
The best part about GPT-3 is that users can access GPT-3 for free using $5 free trial tokens. To get the idea of the model and decide, whether it’s worth it or not.
Apart from this, above we have mentioned how you can measure token usage in GPT-3. Along with factors that might impact the cost of using GPT-3.
