Breaking the Bank: GPT-4.5 Preview Costs $150 per Million Tokens!
Ivana Tilca
Lead Manager @ Allata | Microsoft MVP in Artificial Intelligence | Technology Advocate I Speaker I World Traveler
The pricing and performance of GPT-4.5-preview have sparked quite a bit of buzz, not only for its capabilities but also its cost. Here’s a breakdown:
Why Does GPT-4.5 Cost So Much?
1?? Limited Availability: Being labeled as a "research preview," GPT-4.5 isn't widely accessible yet. This exclusivity naturally impacts its pricing.
2?? Advanced Features: The model showcases enhanced creativity, fewer hallucinations, and other significant improvements. These upgrades demand substantial computational resources, contributing to higher costs.
Pricing Overview
Developers face a steep cost:
This makes it a considerably high investment, especially when compared to previous models like GPT-4. Curious about the detailed pricing? Her is a simple table that can explain it.
OpenAI provides more information on their official pricing page.
Model Comparisons
To evaluate how GPT-4.5 performs compared to earlier iterations or competitors, Andrej Karpathy conducted an interactive comparison via X (formerly Twitter). Participants voted on different scenarios with surprising outcomes:
To see the questions click here. Link to X post: https://x.com/karpathy/status/1895213020982472863
? Question 1: GPT-4.5 is A; 56% of people prefer it.
? Question 2: GPT-4.5 is B; 43% of people prefer it.
? Question 3: GPT-4.5 is A; 35% of people prefer it.
? Question 4: GPT-4.5 is A; 35% of people prefer it.
? Question 5: GPT-4.5 is B; 36% of people prefer it.
Personal Thoughts
Having tested it myself, I can confidently say GPT-4.5-preview stands out for its creativity. While the cost may seem high, its performance truly feels groundbreaking and sets a new standard for AI models.
Let's compare GPT-4 vs GPT-4.5 with a simple prompt.
import openai
openai.api_key = "your-api-key"
def test_prompt(prompt, model):
response = openai.ChatCompletion.create(
model=model,
messages=[{"role": "user", "content": prompt}]
)
return response['choices'][0]['message']['content']
prompt = "Convert 125°F to Celsius."
print("GPT-4 Response:")
print(test_prompt(prompt, "gpt-4"))
print("\nGPT-4.5 Response:")
print(test_prompt(prompt, "gpt-4.5-preview"))
The first result for this prompt is the following:
GPT-4 Response:
The temperature in Celsius is approximately 51.67°C.
GPT-4.5 Response:
To convert Fahrenheit (°F) to Celsius (°C), use the following formula: °C = (°F - 32) \times \frac{5}{9}
Given temperature: 125°F
Applying the formula:
°C = (125 - 32) \times \frac{5}{9}
°C = 93 \times \frac{5}{9}
°C \approx 51.67
Therefore, 125°F is approximately 51.67°C.
Conclusion
Pricing and Performance of GPT-4.5-Preview: Worth the Investment?
The introduction of GPT-4.5-preview has undoubtedly sparked considerable discussion regarding its capabilities and associated costs. Although the high pricing can be attributed to its limited availability and advanced features, the model stands out for its creativity and reduced hallucinations. These improvements demand significant computational resources, justifying the elevated costs. While developers may find the investment steep, the performance of GPT-4.5-preview sets a new benchmark in the AI landscape, making it a noteworthy consideration for those seeking cutting-edge technology.