GPT-5 is here! Should you care?

What’s Actually New in GPT-5

I won’t bore you with every spec, you can read the model card for that. Here’s what matters if you’re running a business or building products:

• Better reasoning: Handles complex, multi-step prompts with fewer errors.

• Improved context memory: Keeps track of longer conversations and nuanced instructions.

• Safer defaults: Fewer off-brand or risky completions, thanks to “safe completions” work.

• Cheaper and faster: Yes, both. And I suspect that the trend will continue.

I tried doing a few zero-shot prompts and it is clearly better than its predecessor when it comes to coding tasks or doing things like building simple games or creating a website.

You can now execute any frontend code from within ChatGPT itself!

Impact: API Users vs ChatGPT-Only Users

If you’re using the API:

• This can be a drop-in upgrade. Just point your endpoint to GPT-5, and you’ll see better reasoning and stability without rewriting anything.

• Fewer “prompt gymnastics” to get the model to behave.

• Lower cost per run and potentially fewer runs thanks to more accurate first outputs.

• It's worth testing Nano/Mini for preprocessing or summarization to optimize spending.

If you’re only using ChatGPT:

• Day-to-day queries just got smarter. You get better follow-through and fewer nonsensical tangents.

• Document analysis, research, and content ideation are noticeably faster and more coherent.

• If you’ve been curious about AI agents via no-code tools (Zapier, Make, n8n), GPT-5 is a stronger base model to start with.

The Prompting Shift

This is where OpenAI’s own GPT-5 Prompting Guide comes into play:

• Think in capabilities, not answers: Ask GPT-5 to plan, reason, and decide — not just reply.

• Use chain-of-thought sparingly: GPT-5 is better at internal reasoning; over-engineering your prompts can sometimes get in the way.

• Leverage the Prompt Optimizer: If you’re migrating from 4.x, run your prompts through it—it’ll surface issues you didn’t know you had.

I have a prompt that I use to classify companies based on their scrapped homepage (via Firecrawl), and it gave me some really nice suggestions.

Let me know what other interesting new use cases you find!
Go use the prompt optimizer!

Opportunities This Unlocks

• AI agents that can actually follow multi-step plans without babysitting.

• More personalized, on-brand AI interactions for customers.

• Cost-effective product features. Things that didn’t make sense with GPT-4 Turbo’s pricing suddenly become viable.

• Richer multi-modal possibilities if/when GPT-5’s image and voice capabilities are fully unleashed.

For now, I tried using the image capabilities and it seems that they have not improved. But I’m hoping that will change soon!

Upcoming Experiments

Over the next couple of weeks, I’ll be testing GPT-5 in:

• Internal Slack-based AI agents that automate project tracking and reporting.

• AI automations around sales and marketing.

• Generative product features for clients in healthcare and finance.

• Customer support flows to see if we can safely reduce human handoffs.

I’ll share what works (and what breaks) in the next edition of Brew. Build. Breakthrough.

I was away for the long weekend, and the GPT-5 bombshell just dropped when I was about to leave. Almost made me take my laptop with me :)

Your turn, whether you’re using GPT-5 through the API or just in the ChatGPT app.
What’s the first thing you’re trying with it?

Reply to this email. I read every one.

And if you’re new here:

About me

Brew. Build. Breakthrough.

Karan Shah
Founder & CEO, SoluteLabs
Building AI-native products before it became cool.