Is the New ChatGPT 4.5 Worth Using for Chat and Writing?

TL;DR

GPT-4.5 is useful for empathetic writing, idea development, and document analysis, but the reviewer found it was not worth upgrading to the $200 monthly Pro plan solely for early access. It still fabricated information, offered limited gains over GPT-4o in some tests, and had API pricing of $75 per million input tokens and $150 per million output tokens. Read on for the specific strengths, failures, and comparisons behind that conclusion.

Transcript

open AI just released GPT 4.5 and I've had almost a full day now playing around with it testing it so I wanted to share some of my results with you in this video I compared it against some other models including the new clae 3.7 model as well as the old chat GPT model too and as far as who has access to GPT 4.5 right now and by the way this is stil... Read More

Key Insights

• GPT-4.5 is a non-reasoning model designed primarily for chat, broader knowledge, writing, and humanlike interaction. OpenAI described it as its largest and best chat model, while the reviewer noted that benchmark performance falls short of reasoning systems such as o1 and DeepSeek R1.
• Early GPT-4.5 access was limited to the $200 monthly Pro plan when the review was recorded. The reviewer said Plus and Teams access was expected the following week, with Education and Enterprise access expected one week later, and advised viewers not to upgrade solely for access.
• GPT-4.5 API pricing was presented as $75 per million input tokens and $150 per million output tokens. The reviewer considered this pricing unusable for typical application development and automation work, especially compared with the less expensive models included in the requested comparison.
• GPT-4.5 can still hallucinate detailed information about nonexistent subjects. When asked about an invented Orange Cream mango, it fabricated a Florida origin, physical characteristics, and other details, both without explicit search and after the reviewer enabled search.
• Search access does not guarantee factual accuracy in GPT-4.5. An initial pricing response used an unreliable source and reported an incorrect figure, while a follow-up request to double-check the answer retrieved accurate information from OpenAI and Anthropic sources.
• GPT-4.5 produces strong emotionally sensitive writing. Its message to employees remaining after layoffs was judged sincere, empathetic, appropriately formatted, and well sized, showing a subtle improvement in tone that may be most noticeable to frequent users.
• GPT-4.5 gives clear and practical everyday guidance, but its advantage over GPT-4o may be small. Both models produced useful laptop battery recommendations, and GPT-4o offered a more comprehensive response with 17 tips organized into categories.
• GPT-4.5 can expand broad ideas into detailed plans and analyze documents with little context. It developed an AI business insight platform through target audience, functionality, technology, monetization, and benefits, and it identified termination, exclusion, and compelled-disclosure concerns in a boilerplate NDA.

Questions & Answers

Q: Is GPT-4.5 worth upgrading to the Pro plan for?

The reviewer said not to upgrade to the $200 monthly Pro plan solely for early GPT-4.5 access. Although the model performed well at empathetic writing, idea development, and document analysis, its improvements over GPT-4o were sometimes subtle, and it still fabricated facts.

Q: What is GPT-4.5 best at according to the tests?

GPT-4.5 performed strongly in empathetic communication, polished writing, practical explanations, business ideation, and document analysis. It drafted a sincere message after layoffs, developed an AI business concept, and identified potential issues in a boilerplate NDA.

Q: Is GPT-4.5 a reasoning model?

No, GPT-4.5 was presented as a non-reasoning model designed primarily for chat, broader knowledge, writing, and humanlike interaction. OpenAI called it its largest and best model for chat, while the reviewer noted that its benchmarks fell short of reasoning models such as o1 and DeepSeek R1.

Q: Does GPT-4.5 still hallucinate?

Yes. The review showed GPT-4.5 inventing a Florida origin, physical characteristics, and other details for a nonexistent Orange Cream mango, even after search was enabled. The displayed comparison reported a 37% hallucination rate for GPT-4.5 versus 61% for another model.

Q: Does web search make GPT-4.5 reliably accurate?

Not consistently in the reviewer's tests. GPT-4.5 initially returned incorrect API pricing from an unreliable source and corrected it only after being asked to double-check, while its search-enabled response still fabricated the Orange Cream mango.

Q: How much does the GPT-4.5 API cost?

The pricing presented in the review was $75 per million input tokens and $150 per million output tokens. The reviewer considered those prices impractical for developing AI applications or automations.

Q: How does GPT-4.5 compare with GPT-4o for everyday advice?

GPT-4.5 produced 10 clear tips for improving laptop battery life, including separate guidance for Windows and Mac. GPT-4o gave similar recommendations but supplied 17 tips organized into categories, so the reviewer found no major GPT-4.5 improvement for that task.

Q: Can GPT-4.5 develop business ideas and analyze documents?

Yes, it expanded an AI-powered business insight platform into a plan covering the target audience, functionality, technology, monetization, and benefits. It also reviewed a boilerplate NDA and raised concerns involving termination, exclusions, and compelled disclosure despite receiving little context.

Summary & Key Takeaways

• GPT-4.5 was presented as OpenAI's largest and best model for chat, but not as a reasoning model. Its improvements appeared most clearly in conversational tone, empathy, writing quality, idea development, and detailed responses. The reviewer found these gains useful but often subtle compared with the capabilities already available in GPT-4o.

• Reliability remained a serious weakness during testing. GPT-4.5 initially supplied incorrect API pricing information and later invented a detailed description of a nonexistent Orange Cream mango. Turning on search did not prevent that fabrication, while Perplexity correctly stated that no specific information about the invented mango was available.

• The model performed well when drafting a message to employees after layoffs, suggesting laptop battery improvements, expanding an AI business idea, and identifying potential NDA issues. However, its high API cost, continued hallucinations, and limited advantage over GPT-4o led the reviewer to discourage paying for Pro access solely to use it early.