GPT-4 (Custom GPT)
Last reviewed
May 9, 2026
Sources
10 citations
Review status
Source-backed
Revision
v3 · 2,332 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 9, 2026
Sources
10 citations
Review status
Source-backed
Revision
v3 · 2,332 words
Add missing citations, update stale details, or suggest a clearer explanation.
| GPT-4 (Custom GPT) | |
|---|---|
| Information | |
| Name | GPT-4 |
| Platform | ChatGPT |
| Store | GPT Store |
| Model | GPT-4 |
| Category | ?????? |
| Description | The most basic version of the GPT-4 model, without any plugins. |
| Developer | XIANG SHENG |
| OpenAI URL | https://chat.openai.com/g/g-rd1ZJ0jtZ-gpt-4 |
| GPT ID | g-rd1ZJ0jtZ |
| Chats | 739,000 |
| Free | Yes |
| Available | Yes |
| Updated | 2024-01-23 |
GPT-4 is a Custom GPT for ChatGPT published in the GPT Store by an independent builder credited as XIANG SHENG. Despite its name, this Custom GPT is not an official OpenAI product. It is a third-party configuration that wraps the underlying GPT-4 model with no custom instructions, no uploaded knowledge files, and no enabled tools, presenting itself in the store's own description as "the most basic version of the GPT-4 model, without any plugins." [1]
The entry attracted notable usage during the early months of the GPT Store, accumulating around 739,000 conversations by January 2024, which placed it among the more frequently visited GPTs created by individual users in that period. Its appeal appears to stem largely from its name, which directly matches the underlying large language model that powers it, rather than from any unique customization. The Custom GPT is identified by the URL slug g-rd1ZJ0jtZ, originally hosted at chat.openai.com and later automatically redirected to chatgpt.com after OpenAI consolidated its consumer surface onto the chatgpt.com domain in 2024.
Custom GPTs are user-created variants of ChatGPT that combine a base OpenAI model with a builder-supplied system prompt, optional reference files, optional tool access (such as web browsing, DALL-E image generation, or the code interpreter), and optional Actions that connect to external APIs. They were first introduced by OpenAI on November 6, 2023, at the company's inaugural DevDay conference, where Sam Altman demonstrated a tool called the GPT Builder for creating them through natural-language conversation. [2]
The GPT Store itself opened to the public on January 10, 2024, after several weeks of delay during which OpenAI added review processes to handle the volume of submissions. [3] OpenAI reported that more than three million Custom GPTs had been created by the time the store launched, with roughly 159,000 of those visible in the public directory. The GPT-4 Custom GPT covered in this article was published into that initial wave and was last updated on January 23, 2024, less than two weeks after the store became available.
The builder name displayed on the GPT's profile is XIANG SHENG, written in roman characters with a space between the two parts. OpenAI requires GPT publishers to verify their identity before listing in the GPT Store, either through a domain-verified website or by displaying their verified social profile or builder name. The name shown on this entry corresponds to the verified builder name attached to the account. No further public information about the builder, such as a personal website, organizational affiliation, or other published GPTs, is associated with this profile in publicly indexed sources.
Because OpenAI did not register "GPT" as a trademark in time, and because the United States Patent and Trademark Office formally refused that registration in February 2024 on the grounds that the term is "merely descriptive," individual builders were free to use the bare model name as a Custom GPT title without violating store policy. [4] Several Custom GPTs in the early store carried names like "GPT-4," "GPT-4 Vision," "GPT-4 Turbo," or "ChatGPT," without being made by OpenAI itself. The XIANG SHENG entry is one of the most heavily used examples of this pattern.
Unlike most popular GPTs in the store, this entry contains essentially no customization. The only system prompt visible to it is the boilerplate scaffold that OpenAI's GPT Builder adds automatically to every new GPT. The full text of that scaffold reads:
You are a GPT, a version of ChatGPT that has been customized for a specific use case. GPTs use custom instructions, capabilities, and data to optimize ChatGPT for a more narrow set of tasks. You yourself are a GPT created by a user, and your name is GPT-4. Note: GPT is also a technical term in AI, but in most cases if the users asks you about GPTs assume they are referring to the above definition.
No additional instructions appear after the standard scaffold. There are no defined conversation starters, no uploaded reference files (the Knowledge field is empty), and no enabled tools or Actions. Web browsing, image generation, and the code interpreter are all disabled by default in this configuration, although a user with a paid subscription would still be routed to the underlying GPT-4 model when chatting with it.
The practical effect is that conversing with this Custom GPT is almost identical to conversing with default ChatGPT under the GPT-4 model, with two differences. First, the GPT identifies itself by the name "GPT-4" if asked. Second, because Custom GPT memory and conversation history are kept separate from the regular ChatGPT thread list, conversations started inside this GPT are filed under the GPT's own thread folder rather than the main chat history.
The GPT Store displays an aggregate conversation counter on each public GPT page, rounded to the nearest thousand. As of late January 2024, the entry showed approximately 739,000 conversations, which made it one of the most-engaged-with non-featured GPTs published by an individual builder rather than by a brand or by OpenAI itself. For context, store-wide leaders such as Grimoire (a coding helper) and Consensus (a research assistant) reached one to three million conversations during the same window. [3]
The high traffic on a Custom GPT with no custom instructions is widely interpreted in community discussions as the result of users searching the store for "GPT-4" expecting to find an official OpenAI listing, and clicking on the closest-matching name. Because OpenAI does not directly publish the base GPT-4 model as a Custom GPT in the store, this third-party entry sits near the top of name-based search results.
| Metric | Value |
|---|---|
| GPT slug | g-rd1ZJ0jtZ |
| Builder | XIANG SHENG |
| Underlying model | GPT-4 |
| Conversations | ~739,000 |
| Custom instructions | None beyond default scaffold |
| Knowledge files | None |
| Web browsing | Disabled |
| DALL-E image generation | Disabled |
| Code interpreter | Disabled |
| Actions | None |
| Conversation starters | None |
| Last updated | 2024-01-23 |
| Cost to use | Free for ChatGPT Plus subscribers |
When the GPT-4 Custom GPT was published in January 2024, the model selector for new Custom GPTs defaulted to the GPT-4 base model. GPT-4 was first released by OpenAI on March 14, 2023, as a large multimodal model capable of accepting both text and image input. Its training data cutoff is generally given as April 2023, though earlier snapshots used cutoffs in 2021 and early 2022. [5]
Following the May 13, 2024 release of GPT-4o, OpenAI progressively shifted Custom GPTs onto the new flagship model. By default, GPTs that had been authored against GPT-4 began routing through GPT-4o for paid users, and OpenAI later opened limited Custom GPT access to free-tier users via GPT-4o. [6] As a result, although this Custom GPT is named "GPT-4" and was originally configured to call the GPT-4 model, the actual model serving any individual conversation has shifted over time as OpenAI has retired or replaced base models. The Custom GPT itself does not pin a specific model checkpoint, so its behavior in 2025 differs from its behavior in early 2024.
Using this Custom GPT does not cost anything beyond a normal ChatGPT subscription. At launch, the GPT Store was open only to ChatGPT Plus, Team, and Enterprise subscribers, since those were the tiers that included access to GPT-4 at the time. [3] Free users could neither browse the store nor create their own GPTs.
OpenAI broadened access in May 2024, when it announced that free-tier users would be able to use Custom GPTs from the store within the daily message cap on GPT-4o, although they still could not create their own. [6] Building Custom GPTs has remained restricted to paid plans throughout 2024 and 2025.
This entry has been used by AI commentators as a working example of two structural issues with the early GPT Store.
The first issue is name confusion. Several writers covering the GPT Store launch noted that because OpenAI had not reserved its own model names as Custom GPT titles, third-party builders could publish minimal wrappers under names such as "GPT-4" or "ChatGPT" and accumulate large traffic from users who assumed they were using an official product. [7] The XIANG SHENG GPT-4 entry is frequently cited as the canonical example of this dynamic.
The second issue is low-effort listings. Reporters from outlets including The Register and TechCrunch documented the rapid arrival of "copycat" and minimal-customization GPTs in the store's first weeks, raising concerns about whether OpenAI's review pipeline was filtering out duplicates and trivial wrappers. [7] [8] A GPT named after the underlying model with no instructions, no knowledge files, and no tools is a textbook instance of the pattern those reports described.
A more academic look at the question came from research published in 2025 that audited the GPT Store for policy compliance. The researchers found that automated and human moderation pipelines were unable to scale to the size and diversity of the store, and that low-quality or near-duplicate GPTs persisted in public listings well after launch. [9] OpenAI later began removing many such entries during sweeps that prioritized quality over quantity, and the GPT-4 Custom GPT covered here is no longer reachable at its original chatgpt.com URL as of 2025.
Despite, or perhaps because of, its lack of any unique features, this Custom GPT is historically significant for three reasons.
First, it served as one of the most concrete demonstrations that the GPT Store at launch did not enforce naming exclusivity around OpenAI's own product names. That observation contributed to broader public discussion in 2024 about whether "GPT" could be trademarked at all, a discussion eventually resolved when the United States Patent and Trademark Office refused OpenAI's registration application on February 8, 2024. [4]
Second, it is a frequently cited example in tutorials and commentary about how Custom GPTs work, since its system prompt exposes the unmodified default scaffold that OpenAI's GPT Builder injects into every new GPT. Comparing this scaffold against the system prompts of more elaborate GPTs makes it possible to see exactly which lines OpenAI adds automatically and which lines are the work of the builder. [10]
Third, the conversation count attached to this entry, in the high six figures, illustrates the long tail of casual usage in the GPT Store: a GPT can attract substantial traffic without ever being formally featured by OpenAI, simply by sitting at a fortunate intersection of search and naming.
A number of other Custom GPTs in the store similarly carry names that mirror official OpenAI products or branding. Examples documented elsewhere on this wiki and in third-party directories include various entries titled "GPT-4," "GPT-4 Turbo," "GPT-4 Vision," "ChatGPT," and "DALL-E," each with different builders and varying degrees of actual customization. The GPT-4 by XIANG SHENG is one of the highest-trafficked among them.
See also: Custom GPTs, GPT Store, GPT-Builders' Assistant, 22.500 Best Custom GPTs, GPT-4, ChatGPT, and OpenAI.
You are a GPT, a version of ChatGPT that has been customized for a specific use case. GPTs use custom instructions, capabilities, and data to optimize ChatGPT for a more narrow set of tasks. You yourself are a GPT created by a user, and your name is GPT-4. Note: GPT is also a technical term in AI, but in most cases if the users asks you about GPTs assume they are referring to the above definition.
None defined.
None uploaded.
None configured.
Because this Custom GPT contains no instructions or tools beyond the default scaffold, interacting with it is functionally similar to interacting with the underlying GPT-4 model in default ChatGPT. To open it, paid subscribers can navigate to the original GPT URL above (or its current chatgpt.com redirect, if it remains live), then send any prompt that they would normally send to GPT-4. The model will respond as it would in a standard ChatGPT session, except that it will identify itself by the name "GPT-4" rather than "ChatGPT" if asked.
Because the Custom GPT is unconstrained, the prompts that work well with it are the same general-purpose prompts that work with default GPT-4: summarization requests, code questions, drafting tasks, brainstorming, translation, and similar open-ended use cases.
No public example transcripts are associated with this entry. Conversation logs inside Custom GPTs are private to the user account that initiated them.