At the end of February, OpenAI released ChatGPT-4.5, its latest update to the ChatGPT product. It’s been nearly a year since OpenAI last introduced a classic Large Language Model (LLM), the May 2024 release of GPT-4o. That update was a significant step-change in performance, offering faster speeds, better reasoning, and lower costs.
Since then, anticipation for the next model has been growing. Would the next version of ChatGPT build on that momentum? Would it deliver meaningful improvements? Or would it confirm what everyone suspects, that we have hit some limits in Generative AI scaling.
Now that the model is here, the response has been mixed. One anonymous expert claims that “ChatGPT-4.5 is lemon!”. But could this lemon model be the best yet for marketing and comms use cases?
Let’s find out.
While OpenAI’s latest release brings some refinements, it is widely seen as one of the company’s most inefficient models yet.
The model demanded massive computational resources to develop, making it significantly more expensive to train than GPT-4o. Higher development costs aren’t unusual, but they typically come with major performance improvements. That’s not the case here.
GPT-4.5 falls short on key benchmarks, especially in math and science reasoning. While it edges out GPT-4o, it lags behind DeepSeek V3, OpenAI’s own Chain of Thought models (like o3-mini), and xAI’s Grok 3. And when it comes to coding tasks, early testing suggests it struggles with complex problems, making it a poor choice for developer as well.
Then there’s the pricing problem. Using GPT-4.5 through OpenAI’s API costs $75 per million input tokens and $150 per million output tokens, a massive jump compared to competitors. DeepSeek V3, for example, charges as little as $0.07 per million input tokens and $1.10 per million output. In an industry where new models are becoming cheaper and more efficient, GPT-4.5 is moving in the opposite direction, making it one of the most expensive models on the market.
The rollout of ChatGPT-4.5 has sparked a flurry of discussions among users, particularly on platforms like Reddit. While some early adopters praised the model's initial responsiveness, deeper interactions unveiled notable inconsistencies. A recurring theme is GPT-4.5's struggles with prompt adherence; users report that, despite detailed instructions, the model often overlooks key details, more so than its predecessor, GPT-4o. Additionally, there's a perception that the model's performance declines over extended sessions, potentially pointing to resource allocation challenges within OpenAI.
OpenAI CEO Sam Altman has openly addressed the challenges accompanying the ChatGPT-4.5 rollout. He revealed that the company is currently "out of GPUs," necessitating a staggered release of the new model. It’s also important to note that GPT-4.5 has only been released as a research preview, which may explain some of the inconsistencies reported by early users. OpenAI often fine-tunes its models after launch, meaning performance could improve over time—but for now, the experience has been far from seamless.
ChatGPT-4.5 might struggle with math and coding, but let’s be honest, most marketing and comms professionals aren't known for their complex calculations or programming expertise. What they do need is compelling copy, creative ideas, and content that resonates with audiences, all areas where language models excel.
One of the most notable improvements is how naturally it writes. In my testing, I found sentences flowed more smoothly than with GPT-4o, and the model was capable of nuanced, well-structured writing across a range of topics. This makes it better suited for long-form content, brand storytelling, and social media copy.
It also shows a stronger grasp of tone and style. Compared to GPT-4o, GPT-4.5 was better at adjusting to my stylistic preferences, modulating tone, and capturing emotional nuance, all of which help save time in editing and recrafting content to fit a specific brand’s tone of voice.
Creativity has also taken a step forward. In my experience, GPT-4.5 performed particularly well in brainstorming, structuring thought leadership content, and generating creative variations of marketing materials. For teams that rely on AI for early-stage content development, this update offers greater flexibility and originality than previous versions.
Beyond writing, GPT-4.5 demonstrates more human-like interactions and aesthetic intuition. It has an improved ability to detect sentiment and adjust responses accordingly, which makes it particularly effective for customer service and support.
Finally, GPT-4.5 has a lower hallucination rate, reducing the risk of inaccurate or misleading information.
With the release of GPT-4.5, we may be witnessing the end of OpenAI’s classic model approach. The company has already hinted that ChatGPT-5 will move beyond the standard LLM approach, focusing more on synthesizing multimodal capabilities and specialized reasoning models. If that’s the case, GPT-4.5 may be the last of OpenAI’s classic models before it shifts to this new approach.
But beyond OpenAI’s roadmap, GPT-4.5 also raises bigger questions about the future of AI scaling. As we covered in our last issue of the newsletter, AI companies are diverging into two camps: compute maximalists, who believe in bigger models trained with massive GPU resources, and efficiency-first approaches that optimize performance with smaller, more specialized architectures.
GPT-4.5 seems to validate the idea that scaling alone may no longer be the answer. Despite requiring massive compute resources, its improvements over GPT-4o are relatively minor. This has major implications for Artificial General Intelligence (AGI), if bigger models aren’t delivering exponential gains, AI developers will need new breakthroughs in reasoning, efficiency, and adaptability rather than just larger datasets and more GPUs.
For marketing and communications professionals, however, the bigger question is whether GPT-4.5 is worth the cost. While it delivers clear improvements in writing fluency, persuasion, and sentiment detection, its high API price makes it difficult to justify, especially when competitors offer similar results at a fraction of the cost.
Instead, the real value of GPT-4.5 may come from its use inside ChatGPT itself. Rather than relying on expensive API calls, marketing teams may find it more practical to use GPT-4.5 directly within the ChatGPT interface for content creation, messaging, and brainstorming.
If this is indeed OpenAI’s last classic model, it’s not ending on a high note.
For more insightful content, subscribe to our newsletter or check back next week for the latest deep dives into Generative AI, industry trends, and what’s next in the world of artificial intelligence.
