ChatGPT 5 - What's ACTUALLY New?
🕓 Read Time: ~3 minutes
While I was on my summer break, OpenAI rolled out ChatGPT-5, so I’ve been busy catching up. But I don't test new models by reading specs (I skim them, at best). I test them hands on, with my most frequent workflows.
Anyways, first things first: my initial thoughts on ChatGPT 5...
The biggest win? They finally cleaned up the model chaos. You no longer have to guess which version to use. It’s now just GPT-5, and it adapts between "fast" responses or "thinking" ( = reasoning), depending on what you need. You can either manually select it, or let ChatGPT decide what's best.
I am not yet sold on either way, to be honest, though I am leaning towards manual selection, depending on my needs.
Other than that, here’s a quick look at what else is different:
-
Fewer hallucinations: Up to 80% fewer factual errors in complex tasks (says OpenAI, I cannot confirm that, yet).
-
More “you”: Built-in personalities, improved voice features, and long-term memory.
-
Massive context window: Can now hold entire books or long threads (256k+ tokens).
-
Multimodal power: Handles text, images, audio, and video.
Sounds impressive? It does, but what really matters to me is whether it makes my life easier... or not. So the real work begins now: testing the new model and all its features on those use cases that matter most in my daily work.
Here are the 5 tests I’m planning:
-
Deep research: does it use more sources, and is the output more accurate?
-
Authenticity of text generation: does it sound human and natural, or still “AI-polished”? Is it really better than 4o?
-
Synthesis of different documents: how well does it handle large amounts of text?
-
Conceptual work: can it structure a learning curriculum or program?
-
Multimodality in action: testing text, images, audio, and video features.
Now I'd love your input: Which of these tests should I write about first? Reply with your top pick (or suggest your own).
Key Takeaway:
GPT-5 may be the most capable model yet, but specs only tell part of the story. The real test is whether it delivers in your day-to-day work.
Til next time!
Elena
P.S.: I’ll share my first round of results in an upcoming edition. Want to shape the test plan? Hit reply and tell me what you’d like me to explore.
