GPT 5 Is Here!

In this edition: an overview and exploration of the latest GPT model from OpenAI.

This is Q&AI, our blog series aimed at keeping you in the know on updates in the rapidly evolving world of AI. Sometimes, these will be quick updates on new developments in the field. Sometimes, they’ll be tips on tactics, features, or functionality.

If you haven’t met me yet, hi: I’m Jen Taylor, CI’s Director of AI Strategy & Implementation, and your (very human) AI BFF. AI is moving at the speed of light, so I’m here to let you know what matters most now. 

Q: WHAT’S THE DEAL WITH GPT 5?

A: ChatGPT 5 started rolling out (slowly) on August 7.  It’s been a real journey. As expected, this is a unified model, meaning users no longer need to choose between a “fast” model or a “reasoning” model. It dynamically adjusts how deeply it thinks based on your prompt, streamlining the user experience significantly.

Sam Altman, OpenAI’s CEO, described the new model experience as “like talking to an expert” and emphasized that it can now “do stuff”, a nod to its enhanced capabilities and tool integration.

Key Enhancements, according to OpenAI:

• Fewer hallucinations and more factual accuracy
• Reduced misleading or deceptive responses
• Improved writing quality across tone and formats
• Stronger coding capabilities
• Artifacts-style functionality; OpenAI demoed new features that mirror Anthropic’s “Artifacts,” allowing users to interact with code outputs or documents in real time
• More natural-sounding voice output
• Integration with Gmail and Google Calendar (rolling out next week, CI Team Account Users Please Do Not Connect)
• A deeper focus on health-related use cases. Sam specifically noted that future models (like GPT-5) are being developed to provide more helpful, trustworthy support around health questions and challenges. This was highlighted as one of the most important applications for the technology.

One of the biggest challenges of the model is that while it’s a “unified model,” it’s not an actual unified model: it’s more of a router (the user inputs and it decides which mode to route a prompt to). It seems that when this tool is at its best, it’s very impressive. But when it’s not, it can be just…meh. 

Another challenge is the personality of GPT 4o vs. GPT 5. It turns out people have become attached to the feel of 4o and did not enjoy engaging with 5 as much. There was such an initial uproar that OpenAI brought back 4o as an option after deprecating it, and promised to focus on making 5 warmer.

They also added in post-launch, based on the feedback, the option to choose a mode: ‘auto’ where they believe most people will land; ‘fast’ which is most similar to 4o for day to day work; ‘thinking’ which is most similar to o3 for a reasoning model. Pro/Enterprise users also have an option for ‘pro’ which is a separate, more powerful variant for the most complex tasks.

A few additions to prompting this particular model:

• The ability to choose reasoning (as mentioned above) initially was something to include in your prompt; now you can choose in the model-picker.
• Verbosity (the length/depth of response) is tunable. You can tell it in the prompt low, medium, or high.
• GPT 5 takes instructions very literally, compared to 4o, so sloppy or contradictory instructions hurt performance more than before.
• Chain of Thought can be used. Old models kept reasoning hidden, but GPT 5 lets you request recaps of its thought process.
• OpenAI released a prompt optimizer to help update prompts for GPT 5.

Note: this all happened in less than a week!

The general consensus seems to be that GPT 5 is good, but it’s also not the game changer OpenAI talked up. People were hoping for something amazing after waiting 1.5 years for a new release, and they ended up with something in the same general category as other AI tools…meaning the AI race stays wide open!

TALK SOON,
Jen

Have a question for a future edition? Submit it here!