The Mark Cuban Podcast

Episode: New Updates to GPT-5: From Theory to Reality in AI Progress

Date: August 21, 2025
Host: Mark Cuban

Episode Overview

In this episode, Mark Cuban unpacks significant updates to OpenAI's GPT platform, focusing on the return of the model picker, introduction of new features for GPT-5, and evolving approaches to AI personality and user customization. Cuban explores how these changes impact users, addresses community feedback, and discusses broader trends in AI model development, including parallels with competitors like Anthropic.

Key Discussion Points & Insights

1. Return of the Model Picker & Expanded Options

• OpenAI reinstated the model picker due to user demand.
• [01:15] New features allow users more control, letting them choose how GPT-5 responds: "You can now choose between Auto Fast and Thinking for GPT5." (Mark Cuban)
• Old, popular models, previously depreciated or removed, have also made a return.
• The user interface has been simplified after past confusion:

  “It was like, oh, O3 mini, O3 mini Thinking 0440, like 4.5…so many different models. Very confusing. They kind of consolidated it all and just said, okay, it's all chat GPT5…” [02:30]

2. Breakdown of GPT-5 Modes and Subscription Tiers

• Three Primary Modes:
  • Auto: Uses OpenAI’s routing system (now allegedly fixed) to pick the best model for a query.
  • Fast: Prioritizes speed for quick answers.
  • Thinking: Delivers more in-depth, nuanced responses.
  • Pro: Access to the highest-quality and newest models and products, such as advanced video generation, for $200/month.
• Legacy Models: GPT-4O and others are back by default for paid users.
• Toggle for Additional Models: Paid users can enable a setting to access extras like O3, 4.0, 4.1, and 5 Thinking Mini.

3. Community Feedback Drives OpenAI's Decisions

• Returning GPT-4O was direct feedback from users who felt GPT-5 responses lacked warmth and empathy:

  "A lot of people have requested this back because they said it had like a warmer tone...people that use AI for therapy said they wanted this more." [06:10]

• A notable user story involved someone in Ukraine who relied on GPT-4O for support during wartime, illustrating the importance of AI “personality.”
• Commitment from OpenAI CEO Sam Altman for transparency about deprecating models:

  “If they're ever going to depreciate it again in the future, they’ll give people lots of advance notice.” [06:30]

4. Personalization & AI Personality Customization

• OpenAI is actively developing options for users to tailor their AI’s tone and personality.
• Sam Altman tweet highlighted this direction:

  "We are working on an update to GPT5's personality, which should feel warmer than the current personality, but not as annoying to most users as GPT4O. However...we really just need to get a world with more per-user customization of model personality." [13:00]

• Cuban praises the move:

  “I would rather just pick what personality I want...never worried that it's going to be too much one way or another.” [15:00]

• Indicates a future where AI personalities will vary significantly per user, deepening individualization and use-case specificity.

5. Technical Changes and Limitations

• Rate Limits: Users get 3,000 messages/week with GPT-5 Thinking; heavy users get downgraded to mini models after the cap.
• Context Window: GPT-5 Thinking supports up to 196,000 tokens, a substantial capacity, though less than some competitors’ million-token windows.
• Premium Features: GPT-4.5 and additional models require a $200/month Pro subscription due to high compute/GPU costs:

  “The only way to get access to that is to pay $200 a month, which I'm not sure I want to do for that model...it just uses more GPUs, a lot more compute.” [10:00]

6. Broader Industry Context & Comparisons

• Cuban highlights similar trends in Anthropic’s Claude models, where developers sometimes prefer older versions for specific tasks:

  “They came out with a brand new model and coders preferred the older model...So this isn't just unique to OpenAI.” [20:30]

• Suggests the challenge of balancing progress and user attachment to “legacy” AI capabilities is industry-wide.

Notable Quotes & Memorable Moments

• On Model Convergence:
  "They kind of consolidated it all and just said, okay, it's all chat GPT5 and we'll pick which one you can use." [02:45]

• On Subscription Frustrations:
  "It makes me wonder if [GPT-4.5] is not just a better model than GPT5 altogether. And now I'm kind of, you know, mad they took it away from me for allegedly a better model that's worse." [10:50]

• On Future of Personalization:
  “Should OpenAI be the one that makes the decision on tuning the AI model's warmth?...I would rather just pick what personality I want…” [15:00]

• On Real-World Emotional Impacts:
  “She talked to GPT-4O and it was like her friend and therapist...she was like super sad and depressed because GPT-5 wasn't as kind of a model or warm.” [14:00]

• On Industry Parallels:
  “Claude and Anthropics actually went through a very similar thing...This isn't just unique to OpenAI, but it is interesting and it's interesting that OpenAI has walked it back.” [20:45]

Timed Highlights

• [01:15] Introduction of new model picker and explanation of “Auto,” “Fast,” and “Thinking” modes.
• [06:10 – 06:30] Discussion on community pushback for removing GPT-4O and OpenAI's new policy for advance notice.
• [10:00 – 11:00] Concerns about model paywalls and compute costs; Cuban’s personal reaction.
• [13:00 – 15:00] Sam Altman’s pledge for more AI personality customization; discussion of user control over tone.
• [20:30 – 21:00] Comparisons with Anthropic, broader patterns in user-preferred legacy models.

Conclusion

Cuban’s episode provides a critical, user-oriented overview of the recent shifts in OpenAI’s approach to AI model delivery and customization. Featuring user stories, industry analysis, and direct commentary from leadership, this episode captures the pulse of AI’s evolving relationship with its user base—balancing innovation, technical constraints, and deeply personal use cases.