I'm wondering what are the differences in capabilities between these two currently available (current date: September 26, 2025) models (ChatGPT 5 and ChatGPT 4o) in the picker? What are their differences in training, output, and accuracy?
#what is the current difference between chatgpt 5 and chatgpt 4o, when selected from the model picker
cinder shard
quick hornetBOT
Hi @cinder shard. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
Here’s a summary of the current differences between ChatGPT-5 and ChatGPT-4o as selectable models in the ChatGPT model picker (as of September 26, 2025):
GPT-5
- Default Model: GPT-5 is now the default for all logged-in users. It’s described as OpenAI’s flagship model, designed to be both smart and fast, combining the best of previous models into a single auto-switching system.
- Reasoning: GPT-5 features breakthrough reasoning, with the ability to “think” more deeply for complex tasks (e.g., coding, science, data analysis). It can automatically switch between “Chat” (fast) and “Thinking” (deeper reasoning) modes, or you can manually select these modes. There’s also a “Pro” mode for research-grade intelligence, available to Pro and Business users.
- Capabilities: Excels at writing, coding, information synthesis, and even healthcare use cases. It’s designed to deliver more accurate and reliable responses, especially for complex or analytical tasks.
- Speed: Delivers significantly faster responses than previous reasoning models.
- Training Data: No explicit training cutoff is mentioned, but it is implied to be the most up-to-date and advanced model.
- Output: More accurate, reliable, and contextually aware, especially for complex queries. Designed to get the best answer every time.
- Availability: Available to all ChatGPT tiers, with additional manual control for paid users. Canvas and image generation are not available with GPT-5 Pro mode.
- Legacy Models: GPT-4o is now considered a legacy model and is available under the “Legacy models” section for paid usersGPT-5 in ChatGPT.
GPT-4o
- Legacy Model: GPT-4o is now a legacy model, still available in the picker for paid users.
- Multimodal: GPT-4o is OpenAI’s flagship multimodal model, capable of reasoning across text, vision, and audio in real time. It supports features like voice, image generation, and canvas (which are not available in GPT-5 Pro).
- Capabilities: Strong at data analysis, search, image generation, canvas, voice, and deep research. It is especially adept at understanding and analyzing images, charts, and diagrams, and has improved performance in STEM, coding, and visual tasks.
- Training Data: The latest version of GPT-4o in ChatGPT has a knowledge cutoff of June 2024, making it quite current, though GPT-5 is implied to be at least as current or more soUpdates to GPT-4o in ChatGPT (Jan 29, 2025).
- Output: GPT-4o is described as intuitive, creative, and collaborative, with enhanced instruction-following and formatting accuracy. It is also noted for its improved problem-solving in STEM and coding, and for being concise and clear in its responsesImprovements to GPT-4o (March 27, 2025).
- Speed: Fast and responsive, but GPT-5 is positioned as even faster and more accurate for most tasks.
- Availability: Available to all tiers, but as a legacy option for paid users.
Key Differences
- Reasoning & Accuracy: GPT-5 is positioned as smarter and more reliable, especially for complex reasoning and analytical tasks. It can automatically switch to deeper reasoning when needed, while GPT-4o is strong but not as advanced in this area.
- Speed: GPT-5 is faster and more efficient, especially in its default “Auto” mode.
- Multimodality: GPT-4o supports real-time audio, vision, and text, and is still used for Voice mode. GPT-5 (especially in Pro mode) does not support some multimodal features like canvas and image generation.
- Model Status: GPT-5 is the new default and flagship; GPT-4o is now a legacy model, retained mainly for compatibility and specific multimodal use cases.
- Training Data: GPT-4o’s latest version is trained up to June 2024. GPT-5’s training data cutoff is not explicitly stated, but it is implied to be at least as current.
In summary:
GPT-5 is the most advanced, accurate, and fastest model for general and complex reasoning tasks, while GPT-4o remains available for its strong multimodal capabilities (especially voice and vision) and as a legacy option for users who need those featuresGPT-5 in ChatGPT Updates to GPT-4o in ChatGPT (Jan 29, 2025) Improvements to GPT-4o (March 27, 2025).
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet thank you! so, i have been seeing some reports that the model picker is recently quietly designed to select 5 even when 4o is selected by the user. Some reports say that openai has started intentionally labeling chats as 4o but actually using 5 as part of the phasing out process for 4o. Is this true? Is the 4o model that's available as a legacy choice actually the original 4o that was available before 5 was released, or is it a different model (maybe 5?) that has been trained to respond similarly to 40?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
Based on the available knowledge sources, there is no official information confirming that OpenAI is labeling chats as GPT-4o while actually using GPT-5, or that the GPT-4o legacy model in the picker is secretly a different model (such as GPT-5) trained to mimic GPT-4o.
The documentation states that for paid users, GPT-4o remains available as a legacy model in the model picker, and you can manually select it if you wish. When you select GPT-4o from the legacy models section, you are intended to be using the original GPT-4o model, not GPT-5 or a GPT-5 variant. There is no mention of any substitution or silent replacement of GPT-4o with GPT-5 in the model picker. The documentation does clarify that if you open an old conversation that used GPT-4o (or other legacy models) and you do not have legacy models enabled, ChatGPT will automatically switch that chat to the closest GPT-5 equivalent—but this is a visible, intentional switch, not a hidden oneGPT-5 in ChatGPT.
If you have legacy model access enabled, the legacy models (including GPT-4o) will appear in the picker and can be selected for new chats. There is no indication in the official documentation that these are anything other than the original models, subject to the same credits and limits as beforeLegacy Model Access for Business, Enterprise, and Edu Users.
In summary:
The knowledge sources do not support the claim that GPT-4o in the model picker is actually GPT-5 or that OpenAI is mislabeling models as part of the phase-out process. If you select GPT-4o as a legacy model, you are using the original GPT-4o, not a disguised GPT-5. If this changes, OpenAI would be expected to update their documentation accordingly.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet thank you! do you know when openai plans to phase out 4o? if at all/ever?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
As of now, the knowledge sources do not specify an exact date or timeline for the complete phase-out or removal of GPT-4o from ChatGPT. The documentation states that GPT-4o is available as a legacy model for paid users (Plus, Pro, Business, Enterprise, and Edu) if legacy model access is enabled, but there is no mention of a scheduled deprecation or shutdown for GPT-4o in ChatGPT itselfLegacy Model Access for Business, Enterprise, and Edu Users.
However, it is clear that GPT-5 is now the default model, and legacy models like GPT-4o are intended for transitional use. OpenAI typically provides advance notice before fully retiring legacy models, but at this time, there is no official announcement regarding the final phase-out of GPT-4o in ChatGPT. If such a plan is announced, it will likely be communicated through official release notes or help articles.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
why did openai change so much about the interaction style of chatgpt 5 compared to 4o?
@quick hornet why did openai change so much about the interaction style of chatgpt 5 compared to 4o? you mentioned intuitiveness and collaboration- a lot of people have felt like 5 doesnt have these traits. is that true, and if so why did openai change these features? also, what are the current hallucination rates of each model as of 9/26/25?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
OpenAI changed the interaction style of ChatGPT-5 compared to GPT-4o in response to user feedback and ongoing research into making the assistant more effective and approachable. With GPT-5, the default personality was initially described as more reserved and professional, but after feedback that it felt less warm and intuitive than previous models, OpenAI updated GPT-5’s default personality to be “warmer and more familiar.” This means GPT-5 now includes small acknowledgements and more personable interactions, aiming to make conversations feel more approachable without crossing into sycophancy (excessive flattery). OpenAI explicitly distinguishes warmth from sycophancy and is actively working to balance these traits, as detailed in their system card and blog posts on the topic. Teaching models to be warm without being sycophantic remains an ongoing research challenge, and OpenAI continues to iterate on this aspectGPT-5 Updates, Aug 15, 2025.
Regarding intuitiveness and collaboration, OpenAI’s release notes for GPT-4o highlighted improvements in these areas, describing 4o as more intuitive, creative, and collaborative, with enhanced instruction-following and clearer communication. Some users have reported that GPT-5 feels less intuitive or collaborative compared to 4o, and OpenAI’s updates suggest they are aware of these perceptions and are making adjustments to address themImprovements to GPT-4o, Mar 27, 2025.
On hallucination rates:
The knowledge sources do not provide specific, up-to-date quantitative hallucination rates for GPT-5 or GPT-4o as of September 26, 2025. However, OpenAI states that GPT-5 is “our smartest, most reliable model to date, delivering more accurate responses significantly faster than previous reasoning models,” which implies a reduction in hallucinations compared to earlier modelsGPT-5 in ChatGPT. For GPT-4.5 (an earlier model), OpenAI noted expectations of fewer hallucinations, but no explicit rates are given for GPT-4o or GPT-5.
Summary:
- OpenAI changed GPT-5’s style to be more professional at first, then made it warmer in response to feedback.
- GPT-4o was noted for intuitiveness and collaboration; OpenAI is working to bring these traits to GPT-5.
- No specific hallucination rates for GPT-5 or GPT-4o are published in the current knowledge sources.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet do you know how soon the changes might roll out for 5 to become more intuitive, creative and collaborative, and if they are also going to try to re-instate the enhanced instruction-following and clearer communication? do you know how progress is currently going on this and when these changes are expected?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
I'm sorry, but the knowledge sources do not provide any specific timeline or expected rollout date for updates to make GPT-5 more intuitive, creative, or collaborative, nor do they detail when (or if) enhanced instruction-following and clearer communication—traits highlighted in GPT-4o—will be reinstated in GPT-5.
The most recent updates mention that OpenAI is aware of user feedback regarding GPT-5’s interaction style and has already made adjustments to make the default personality warmer and more familiar, with ongoing research to further improve the model’s conversational qualities. However, there is no explicit information about the progress or schedule for additional changes aimed at matching or surpassing GPT-4o’s intuitiveness, creativity, collaboration, or instruction-following abilities. OpenAI states that more updates are coming soon, but no concrete dates or feature lists are providedGPT-5 Updates, Aug 15, 2025.
If OpenAI announces further improvements or a timeline, it will likely be communicated through official release notes or help articles.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet when was the most recent announcement or update released about their plans to restore the intuitiveness, creativity, collaboration and enhanced instruction-following abilities, and/or clearer communication style?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
The most recent update related to OpenAI’s plans for improving model interaction style—specifically regarding warmth, intuitiveness, and communication—was on August 15, 2025. In this update, OpenAI stated they were making GPT-5’s default personality “warmer and more familiar” in response to feedback that the initial version felt too reserved and professional. They clarified that this update does not increase sycophancy and that teaching models to be warm without being sycophantic is an ongoing research challenge. OpenAI also mentioned that more updates are coming soon, but did not provide a specific timeline or detailed plans for restoring the full range of intuitiveness, creativity, collaboration, or enhanced instruction-following and clearer communication that users associated with GPT-4o. No further or more recent announcements on this topic are present in the knowledge sourcesGPT-5 Updates, Aug 15, 2025.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet were there any other changes announced for 5 since 8/15/25?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
Yes, there have been several additional changes and updates to GPT-5 announced since August 15, 2025:
-
August 12, 2025:
Users can now choose between “Auto,” “Fast,” and “Thinking” modes for GPT-5, giving more control over response speed and depth. Plus users received increased message limits for GPT-5 Thinking, and the context limit for GPT-5 Thinking was set to 196k tokens. Also, legacy models like 4o were made available again in the picker for paid users, with a “Show additional models” toggle in settingsGPT-5 Updates, Aug 12, 2025. -
September 9, 2025:
Updates to ChatGPT Voice mode were announced, but these changes are not specific to GPT-5’s core text modelChatGPT Voice Update, Sep 9, 2025. -
September 17, 2025:
A new “thinking time” toggle was introduced for GPT-5 with Thinking mode, allowing Plus, Business, and Pro users to control how long GPT-5 spends reasoning before responding. Options include Light, Standard, Extended, and Heavy, with different levels of speed and depth. This gives users more flexibility to balance quick answers with deeper reasoning as neededThinking time toggle, Sep 17, 2025.
These are the main changes to GPT-5 announced after August 15, 2025, according to the knowledge sources.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
cinder shard
@quick hornet what were all of the changes to 5 that have been announced since 8/15/25?
quick hornetBOT
Let me check the knowledge base for you again. One moment! :books:
Here are all the changes to GPT-5 that have been announced since August 15, 2025, according to the knowledge sources:
August 15, 2025
- Personality Update:
GPT-5’s default personality was made “warmer and more familiar” in response to feedback that it was too reserved and professional. The update aimed to make interactions more personable without increasing sycophancy. OpenAI noted that teaching models to be warm without being sycophantic is an ongoing research challenge, and more updates are planned for the futureGPT-5 Updates, Aug 15, 2025.
August 12, 2025
- Mode Selection:
Users can now choose between “Auto,” “Fast,” and “Thinking” for GPT-5, providing more control over response speed and depth. - Message Limits:
ChatGPT Plus users now have 3,000 messages/week with GPT-5 Thinking, with additional capacity on GPT-5 Thinking mini after that limit. - Context Limit:
The context limit for GPT-5 Thinking is set to 196k tokens. - Legacy Models:
GPT-4o and other legacy models were made available again in the model picker for paid users, with a “Show additional models” toggle in ChatGPT web settingsGPT-5 Updates, Aug 12, 2025.
September 17, 2025
- Thinking Time Toggle:
A new “thinking time” toggle was introduced for GPT-5 with Thinking mode. Plus, Business, and Pro users can now control how long GPT-5 spends reasoning before responding, with options for Light, Standard, Extended, and Heavy. This allows users to balance speed and depth of reasoning. The setting is currently available on the web and will be coming to mobile apps soonThinking time toggle, Sep 17, 2025.