Table of Contents
- Key takeaways
- What is a GPT-3 chatbot?
- Can you still chat with GPT-3 online?
- Which models replace GPT-3?
- How much cheaper are the current models than GPT-3?
- How to build a GPT-3 chatbot today: three options
- Option 1: Build on the OpenAI API
- Steps
- Cost
- Limits
- Who it fits
- Option 2: Use a no-code chatbot builder
- Steps
- Cost
- Limits
- Who it fits
- Option 3: Use a ChatGPT project or a custom GPT
- Steps for a project
- Steps for a custom GPT
- Cost
- Limits
- Who it fits
- How to choose between the three options
- Can you use a GPT chatbot for customer service?
- Common mistakes when building a GPT-3 chatbot
- Frequently asked questions
- Basics
- What is a GPT-3 chatbot?
- Is ChatGPT the same as GPT-3?
- Is a chatbot the same as GPT-3?
- Access and cost
- Can you still use GPT-3 online?
- Is there a free GPT-3 chatbot?
- How much does it cost to build a GPT chatbot?
- Is the OpenAI API free?
- Building
- How do you build a chatbot with GPT-3?
- Which model should replace GPT-3.5 in an existing chatbot?
- Does a custom GPT have a welcome message field?
- Can a GPT chatbot answer from your own documents?
- Sources
Created by
Do not index
Created time
Sep 22, 2023 09:37 PM
A GPT-3 chatbot today means a chatbot built on one of OpenAI's current models, because OpenAI shut down the GPT-3 models on January 4, 2024 and no product lets you chat with GPT-3 anymore.
TL;DR
- GPT-3 is retired. OpenAI removed ada, babbage, curie, davinci, and text-davinci-003 from its API on January 4, 2024.
- GPT-3.5 is leaving too. gpt-3.5-turbo-instruct shut down on September 28, 2026, and gpt-3.5-turbo is scheduled to shut down on October 23, 2026.
- To chat with an OpenAI model for free, use ChatGPT. The Free plan runs GPT-5.6 Luna and allows unlimited everyday text chats.
- To build a chatbot, pick one of three options: the OpenAI API, a no-code builder, or a ChatGPT project or custom GPT.
- Current models cost far less than GPT-3 did. gpt-6-luna output costs $0.50 per 1M tokens. davinci cost $20 per 1M tokens. That is 40 times more.
Key takeaways
Question | Short answer |
Can you still use GPT-3? | No. The GPT-3 models were shut down on January 4, 2024 |
Is there a free GPT-3 chatbot online? | No. ChatGPT Free is the free option from OpenAI, and it runs GPT-5.6 Luna |
Which models replace GPT-3? | GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna are OpenAI's current flagship API models |
Best option for developers | The OpenAI Responses API with file search, billed by usage |
Best option for a website chatbot without developers | A no-code builder such as SiteGPT, from $59 a month billed monthly |
Best option for personal or team use | A ChatGPT project, or a custom GPT in a Business, Enterprise, or Edu workspace |
What is a GPT-3 chatbot?
A GPT-3 chatbot is a chatbot that sends each user message to OpenAI's GPT-3 language model and returns the text the model writes. GPT stands for Generative Pre-trained Transformer.
GPT-3 was a family of models in the OpenAI API. Its base models were named ada, babbage, curie, and davinci. Developers built chatbots by calling those models through the completions endpoint.
A chatbot and GPT-3 are different things. The chatbot is the product a person types into. GPT-3 was the model behind it. The same chatbot can run on a newer model, and every working "GPT-3 chatbot" now does.
People still search for "GPT-3 chatbot" when they mean any chatbot that runs on an OpenAI model. That is the meaning this guide uses from here on.
Can you still chat with GPT-3 online?
You cannot chat with GPT-3 online anymore, because OpenAI has shut the models down. On July 6, 2023, OpenAI announced the retirement of its older GPT-3 and GPT-3.5 completions models. They stopped working on January 4, 2024.
The models that replaced them have also been retired or scheduled for retirement.
Model | Shutdown date | Replacement OpenAI recommends |
ada, babbage (GPT-3 base models) | January 4, 2024 | babbage-002 |
curie, davinci (GPT-3 base models) | January 4, 2024 | davinci-002 |
text-ada-001, text-babbage-001, text-curie-001, text-davinci-001, text-davinci-002, text-davinci-003 | January 4, 2024 | gpt-3.5-turbo-instruct |
babbage-002, davinci-002, gpt-3.5-turbo-instruct | September 28, 2026 | gpt-5.6-terra |
gpt-3.5-turbo | October 23, 2026 | gpt-5.6-terra |
gpt-4, gpt-4-turbo | October 23, 2026 | gpt-5.6-sol |
Any website that offers a "GPT-3 chat" today is running a different model. If you want to chat with an OpenAI model for free, use ChatGPT. OpenAI's help center says Free users have access to GPT-5.6 Luna and unlimited everyday text chats, subject to abuse-prevention safeguards.
ChatGPT Free also includes web search, file and image uploads, and image creation. Those tools have their own usage limits.
Which models replace GPT-3?
The models that replace GPT-3 are OpenAI's current flagship models: GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. OpenAI's model page recommends them for new work.
Model ID | What OpenAI recommends it for | Input, per 1M tokens | Output, per 1M tokens |
gpt-6-astra | Complex reasoning and coding | $10.00 | $50.00 |
gpt-6.1-sol | A balance of intelligence and cost | $2.00 | $10.00 |
gpt-6-luna | Cost-sensitive, high-volume workloads | $0.10 | $0.50 |
These are standard, short-context prices. Each of the three models has a 1.05M token context window and supports file search, web search, and function calling.
A customer-facing chatbot is a high-volume workload, so gpt-6-luna is the usual starting point. Move to gpt-6.1-sol if answers need more reasoning.
How much cheaper are the current models than GPT-3?
The current models cost a fraction of what GPT-3 cost. OpenAI's deprecations page lists davinci at $20.00 per 1M tokens when it was shut down.
- gpt-6-luna output costs $0.50 per 1M tokens. davinci cost 40 times more.
- gpt-6-luna input costs $0.10 per 1M tokens. davinci cost 200 times more.
- gpt-6.1-sol output costs $10.00 per 1M tokens. That is half the davinci price.
How to build a GPT-3 chatbot today: three options
You build a "GPT-3 chatbot" today by putting a current OpenAI model behind a chat interface. There are three ways to do it, and each one fits a different user.
The three options are compared on four criteria:
- Who can use the chatbot. Website visitors, or only people signed in to ChatGPT.
- Whether it needs code. Someone has to build and host an API chatbot.
- How it is priced. Per token, per seat, or as a flat monthly plan.
- Who keeps the content current. A chatbot repeats whatever its sources say.
No option is best overall. Each one has a "best for" label.
Option | Best for | Needs code | Who can use it | Price to start |
1. OpenAI API | Developers building a chatbot into their own product | Yes | Anyone you build the interface for | Usage-based. gpt-6-luna is $0.10 input and $0.50 output per 1M tokens |
2. No-code builder (SiteGPT) | A customer service chatbot on a website, without developers | No | Website visitors | $59 a month billed monthly, or $39 a month billed yearly. No free plan |
3. ChatGPT project or custom GPT | Personal or team use inside ChatGPT | No | People signed in to ChatGPT | $0 for a project. $50 a month billed monthly for a workspace that can create custom GPTs |
All three options rely on retrieval when the chatbot needs to answer from your content. The model reads the relevant passages at answer time, and its weights do not change. What is RAG explains how that works.
Option 1: Build on the OpenAI API
Best for: developers building a chatbot into their own product.
Building on the OpenAI API means writing code that sends each message to a current model through the Responses API. The file search tool lets the model search your uploaded files before it answers.
Steps
- Create an OpenAI API account and an API key.
- Pick a model. Start with
gpt-6-lunafor a high-volume chatbot.
- Create a vector store and upload your files to it.
- Call the Responses API with the
file_searchtool and the vector store ID.
- Read the answer and the file citations that come back with it.
- Build the chat interface, host it, and add a way for users to reach a person.
Cost
The API bills by usage. Model tokens are billed at the model's rate. File search costs $2.50 per 1,000 tool calls. File storage costs $0.10 per GB per day, and the first GB is free.
Worked example. Assume 10,000 answered questions a month. Assume each question makes one file search call and uses 3,000 input tokens and 300 output tokens. Storage stays under the free 1 GB.
Cost line | On gpt-6-luna today | On davinci at its $20 price |
Input: 30M tokens | $3.00 | $600.00 |
Output: 3M tokens | $1.50 | $60.00 |
Tokens a month | $4.50 | $660.00 |
File search: 10,000 calls at $2.50 per 1,000 | $25.00 | Not available on davinci |
Total a month | $29.50 | $660.00 for tokens alone |
The same token volume costs $4.50 on gpt-6-luna and $660.00 at the davinci price. The difference is $655.50 a month, a drop of more than 99%.
The token counts are assumptions. The totals leave out developer time, hosting, and the chat interface.
Limits
- It needs code. There is no widget, dashboard, or handoff to a person unless you build them.
- GPT-3 era tutorials no longer run. Code that calls
text-davinci-003or the completions endpoint with a GPT-3 model fails.
- The Assistants API was shut down on August 26, 2026. OpenAI names the Responses API and the Conversations API as its replacement.
- Agent Builder is scheduled to shut down on November 30, 2026.
- Fine-tuning is closed to organizations that have not used it before. Plan on retrieval instead.
- Keeping the files current is your job.
Who it fits
The API fits a team with developers that wants the chatbot inside its own app. The guide on how to train ChatGPT on your own data covers API retrieval and fine-tuning in more depth. The guide to a custom AI chatbot covers the build-or-buy trade-offs.
Option 2: Use a no-code chatbot builder
Best for: a customer service chatbot on a website, without developers.
A no-code chatbot builder runs the OpenAI model and the retrieval for you and gives you a chat widget for your website. You add your content, the builder indexes it, and visitors get answers from that content. Visitors do not need a ChatGPT account.
SiteGPT is the worked example here. Its pricing page names the OpenAI models it uses: GPT-4.1-mini and GPT-4.1.
Steps
- Start the 7-day trial. It requires a credit card.
- Create a chatbot and add your content. Sources include website URLs, sitemaps, uploaded files, YouTube, Google Drive, and Notion.
- Write the instructions: what the chatbot covers, its tone, and what it says when the content has no answer.
- Test it with real questions from customers.
- Copy the embed snippet into your website.
- Turn on escalation so visitors can reach a person.
The full walkthrough with screenshots is in how to create an AI chatbot. Content preparation is covered in how to train a chatbot on your own data. If it is unclear whether the bot needs retrieval, written instructions, or fine-tuning, read the guide to chatbot training first.
Cost
SiteGPT plan | Billed monthly | Billed yearly | Messages a month | Chatbots | Content refresh |
Starter | $59 a month | $39 a month ($468 a year) | 4,000 GPT-4.1-mini or 400 GPT-4.1 | 1 | Manual |
Growth | $129 a month | $79 a month ($948 a year) | 10,000 GPT-4.1-mini or 1,000 GPT-4.1 | 2 | Monthly |
Scale | $429 a month | $259 a month ($3,108 a year) | 40,000 GPT-4.1-mini or 4,000 GPT-4.1 | 3 | Weekly, with a daily scan |
Yearly billing on Starter costs $468 a year. Monthly billing costs $708 a year. Yearly billing saves $240, or 34%.
Limits
- There is no free plan. The 7-day trial needs a credit card.
- Message quotas are fixed per plan. A GPT-4.1 message uses 10 times the quota of a GPT-4.1-mini message.
- You cannot pick a GPT-6 model. The pricing page lists GPT-4.1-mini and GPT-4.1.
- Starter has no chat integrations, no API access, and manual content refresh only.
- Removing SiteGPT branding is a paid add-on at $39 a month billed yearly.
- There is no Intercom or HubSpot integration today.
Who it fits
A no-code builder fits a business that wants a website chatbot for customer service and has nobody to build and maintain one. The comparison of AI chatbots for a website covers other platforms in this category. If the budget is zero, see the options in free chatbots for a website.
Option 3: Use a ChatGPT project or a custom GPT
Best for: personal or team use inside ChatGPT.
A ChatGPT project or a custom GPT keeps files and instructions together inside ChatGPT, so every chat starts with the same context. Neither one can sit on a website.
Steps for a project
- Select New project in the ChatGPT sidebar and name it.
- Add files: PDFs, spreadsheets, documents, images, or pasted text.
- Open Project settings and write the project instructions.
- Start a chat inside the project and test it with real questions.
Steps for a custom GPT
New custom GPTs can only be created in ChatGPT Business, Enterprise, and Edu workspaces. Personal accounts on Free, Go, Plus, and Pro cannot create or publish new GPTs.
- In a Business, Enterprise, or Edu workspace, open Explore GPTs in the sidebar.
- Select Create to open the GPT builder.
- Add a name, a description, and conversation starters.
- Write the instructions and upload reference files under Knowledge.
- Test the GPT in Preview, then save it.
Cost
Projects are available to signed-in ChatGPT users, and the number of files depends on the plan. A project on the Free plan costs $0.
A custom GPT needs a Business workspace or higher. A Standard seat costs $25 per user a month billed monthly, or $20 per user a month billed annually. A workspace needs at least 2 seats. The lowest price is $50 a month billed monthly, or $40 a month billed annually ($480 a year).
Limits
- Only people signed in to ChatGPT can use a project or a GPT. A website visitor cannot.
- A custom GPT holds up to 20 knowledge files, each up to 512 MB.
- OpenAI plans to retire custom GPTs. The planned date for Enterprise workspaces is December 11, 2026, and OpenAI expects other plans to follow the same timeline.
- OpenAI recommends moving GPT workflows to Plugins.
Who it fits
A project fits one person or a small team that already works in ChatGPT. A custom GPT fits a team with a Business, Enterprise, or Edu workspace, and that team should plan for the move to Plugins. How to train ChatGPT on your own data lists the file limits for each plan.
How to choose between the three options
Choose between the three options by deciding who will use the chatbot. That one answer rules out most of the list.
If you need... | Pick... | Why |
To chat with an OpenAI model yourself, for free | ChatGPT Free | It runs GPT-5.6 Luna and needs no setup |
The same files and instructions in every chat | Option 3: a ChatGPT project | Files and instructions stay attached to the project |
A shared assistant for a team on ChatGPT Business | Option 3: a shared project or a custom GPT | Workspace members can use it without another tool |
A chatbot on your website that answers customers, with no developer | Option 2: a no-code builder | Visitors need no ChatGPT account, and the widget needs no code |
A chatbot inside your own app | Option 1: the OpenAI API | You control the interface, the model, and the data flow |
The newest OpenAI model on day one | Option 1: the OpenAI API | You choose the model ID yourself |
The lowest cost per answer at high volume | Option 1: the OpenAI API | About $29.50 for 10,000 answers in the worked example, before developer time |
To replace GPT-3 or GPT-3.5 in an existing chatbot | Option 1 with a current model ID | OpenAI names gpt-5.6-terra as the substitute for gpt-3.5-turbo |
Can you use a GPT chatbot for customer service?
You can use a GPT chatbot for customer service if it answers from your own content and hands off to a person. A bare model knows nothing about your prices, policies, or orders.
Three rules matter most:
- Ground every answer. Connect the model to your help pages and files, and tell it to decline when the content has no answer. The guide to chatbot hallucinations covers the other guardrails.
- Give visitors a route to a person. Options 1 and 3 have no handoff unless you build one.
- Test before launch. Run a fixed set of real customer questions and check each answer against the source.
The full list is in chatbot best practices. For a narrow first project, start with an AI FAQ chatbot.
Common mistakes when building a GPT-3 chatbot
The most common mistakes come from following instructions written when GPT-3 was current. Five show up often.
- Following a GPT-3 tutorial line by line. The model names and endpoints in it have been shut down. Use the Responses API and a current model ID.
- Hard-coding a model ID and forgetting it. OpenAI gives at least 6 months of notice before it retires a generally available model. Check the deprecations page on a schedule.
- Building a custom GPT for customers. A GPT lives inside ChatGPT, and OpenAI plans to retire the feature.
- Planning on fine-tuning for knowledge. The fine-tuning platform is closed to new organizations. Facts belong in retrieval.
- Expecting a free website chatbot. ChatGPT Free is free for personal chat. A customer-facing chatbot costs money through the API or a platform.
Frequently asked questions
Basics
What is a GPT-3 chatbot?
A GPT-3 chatbot is a chatbot that uses OpenAI's GPT-3 model to write its replies. GPT-3 has been shut down, so the phrase now describes a chatbot built on a current OpenAI model.
Is ChatGPT the same as GPT-3?
No. GPT-3 was a family of models in the OpenAI API. ChatGPT is OpenAI's chat app, and its Free plan runs GPT-5.6 Luna.
Is a chatbot the same as GPT-3?
No. A chatbot is the product a person talks to. GPT-3 was one language model that could power it. A chatbot can run on any model, or on scripted rules with no model at all.
Access and cost
Can you still use GPT-3 online?
No. OpenAI shut down the GPT-3 models on January 4, 2024. A site that advertises a GPT-3 chat is running a different model.
Is there a free GPT-3 chatbot?
There is no GPT-3 chatbot left, free or paid. ChatGPT Free is the free option from OpenAI. It runs GPT-5.6 Luna with unlimited everyday text chats.
How much does it cost to build a GPT chatbot?
The OpenAI API costs about $29.50 a month for 10,000 answers on gpt-6-luna in the worked example above. SiteGPT starts at $59 a month billed monthly, or $39 a month billed yearly. A ChatGPT Business workspace for custom GPTs starts at $50 a month billed monthly.
Is the OpenAI API free?
No. The API bills per token. gpt-6-luna costs $0.10 per 1M input tokens and $0.50 per 1M output tokens.
Building
How do you build a chatbot with GPT-3?
You cannot build on GPT-3 anymore. Build on a current model instead: call the OpenAI Responses API, use a no-code builder, or set up a ChatGPT project.
Which model should replace GPT-3.5 in an existing chatbot?
OpenAI names gpt-5.6-terra as the substitute for gpt-3.5-turbo. For new work, its model page recommends GPT-6 Luna for high-volume tasks and GPT-6.1 Sol for a balance of intelligence and cost.
Does a custom GPT have a welcome message field?
No. The fields OpenAI documents for a GPT include a name, a description, conversation starters, instructions, and knowledge files. Conversation starters are the closest match. They are example prompts shown when someone opens the GPT.
Can a GPT chatbot answer from your own documents?
Yes. All three options support it. The API uses the file search tool, a no-code builder indexes your pages and files, and a ChatGPT project or GPT reads uploaded files.
Sources
All sources were checked on October 7, 2026.
Written by
