We’ve got big news. You’re today able to use OpenAI's biggest and best model yet - GPT-5 - directly inside Pencil to supercharge your creativity.
From blue-sky brainstorming, to writing sharper scripts, to perfectly tailored demand gen captions, the long-awaited new model is ready to be your copilot.
Getting set up couldn’t be simpler: Log into Pencil, open up the model picker, and select GPT-5.
And that’s it - you’re all set to discover ad creation with the most advanced reasoning, coding, and creative capabilities OpenAI has ever released.
Why this matters for Pencil users
GPT-5 isn’t just a step up — it’s a leap forward. With better reasoning, reduced hallucinations, and faster, more accurate outputs, you can expect:
Sharper copy and concepts: GPT-5 is better at following nuanced briefs.
Fewer rewrites: more accurate responses, first time.
Smarter creative variations: improved ability to adapt ideas across formats.
Why we’re excited
We’ve always promised Pencil users access to the very best generative AI has to offer — and our team moved fast to make it happen. The GPT-5 integration went from announcement to working inside Pencil in just three days.
That speed matters, because it means you can start experimenting with GPT-5 before your competitors even log in.
How to try it
Simply head to your model picker in Pencil and select GPT-5. No extra setup, no delay — just the most advanced model yet, ready to work on your ads.
🧠 Why GPT-5 is such a breakthrough
When OpenAI unveiled GPT-5 this week, they described it as their most capable model yet… and the numbers back it up.
GPT-5 introduces:
Dynamic reasoning mode that kicks in for complex tasks (perfect for multi-step creative briefs).
Higher factual accuracy — up to 45% fewer hallucinations than GPT-4o.
Better context handling for long-form and multi-turn conversations.
Multimodal smarts — improved performance across both text and vision tasks.
For Pencil users, this means your ad copy, creative concepts, and data-driven insights are more likely to be right, relevant, and ready-to-use without extra editing.
🚀 Why this is great for creative work
GPT-5 isn’t just “more accurate” — it’s more creative in a useful way. It:
Handles nuance in brand tone better than any previous model.
Can adapt messaging across formats (social, video scripts, landing pages) with more consistency.
Picks up on subtle brand or product details from your brief and weaves them into ideas without losing context.
Generates more human-like reasoning chains, making creative suggestions feel less generic.
What to try first in Pencil with GPT-5
When GPT-5 lands on Monday, try:
Rewriting your top-performing ad in three new styles to see if GPT-5 can push the creative envelope.
Running complex campaign briefs — see how it connects product benefits to emotional hooks.
Combining GPT-5 with Pencil predictive scoring to see if the higher-quality copy correlates with stronger performance.
The takeaway
With GPT-5, Pencil users now have the most advanced generative model in the world at their fingertips — integrated in record time, ready to help you craft sharper, smarter, and more impactful ads.
From Monday, simply open Pencil, head to your model picker, and select GPT-5. Then… see what’s possible.
Capabilities (graphs created by GPT-5)
GPT-5 delivers state-of-the-art results across multiple domains. In math reasoning (AIME 2025), it reaches 94.6% accuracy without tools. Coding performance is equally strong, scoring 74.9% on SWE-bench Verified and 88% on Aider Polyglot. Its multimodal reasoning (MMMU) comes in at 84.2%, and even on the challenging HealthBench Hard dataset, GPT-5 achieves 46.2% — a meaningful jump for complex, high-context tasks.
GPT-5 Thinking is significantly more truthful than earlier reasoning models. When faced with missing inputs in multimodal tasks, deceptive confident answers drop from 86.7% (OpenAI o3) to just 9%. In real-world production traffic, the overall deception rate falls from 4.8% to 2.1%. For Pencil users, that translates into creative suggestions you can trust.
Accuracy isn’t just about getting high scores — it’s about reducing mistakes. GPT-5 is 45% less likely to hallucinate when web-enabled compared to GPT-4o. And in its “thinking” mode, hallucinations drop by up to 80%. That means fewer made-up facts and more reliable copy, even on complex briefs.