OpenAI announced on Thursday the release of GPT-4.5, its highly anticipated AI model, code-named Orion. As OpenAI’s largest model to date, GPT-4.5 has been trained with more computational resources and data than any of its predecessors.
Thank you for reading this post, don't forget to subscribe!Despite its scale, OpenAI clarified in a white paper that it does not classify GPT-4.5 as a frontier model.
Subscribers to ChatGPT Pro, OpenAI’s $200-per-month plan, can access GPT-4.5 starting Thursday as part of a research preview. Developers using OpenAI’s paid API tiers also have immediate access, while ChatGPT Plus and ChatGPT Team users are expected to receive the model next week.
The AI industry has closely watched the development of Orion, as it serves as a key test of traditional AI training methods. Like its predecessors, GPT-4.5 was built by significantly increasing the computing power and data used in its pre-training phase, a technique known as unsupervised learning. Previous GPT models have seen major performance improvements through scaling, with advancements in mathematics, writing, and coding. OpenAI claims GPT-4.5 has “deeper world knowledge” and “higher emotional intelligence.” However, some benchmarks suggest that its performance gains are leveling off, with newer AI reasoning models from DeepSeek, Anthropic, and OpenAI itself outperforming it in certain areas.
Running GPT-4.5 is also costly. OpenAI acknowledged that its high operational expenses may impact its long-term availability in the API. Developers must pay $75 per million input tokens (about 750,000 words) and $150 per million output tokens—significantly higher than GPT-4o, which costs $2.50 and $10 per million tokens, respectively.
“We’re introducing GPT-4.5 as a research preview to better understand its capabilities and limitations,” OpenAI stated in a blog post. “We’re eager to see how users experiment with it in unexpected ways.”
OpenAI has positioned GPT-4.5 as a specialized model rather than a direct successor to GPT-4o, the primary model powering ChatGPT and OpenAI’s API. While GPT-4.5 supports features like file and image uploads and ChatGPT’s canvas tool, it currently lacks real-time two-way voice capabilities.
On OpenAI’s SimpleQA benchmark, which evaluates models on factual accuracy, GPT-4.5 outperforms GPT-4o and OpenAI’s reasoning models, o1 and o3-mini. OpenAI also claims GPT-4.5 hallucinates less frequently than most AI models, meaning it is less likely to generate incorrect information.
However, OpenAI did not compare GPT-4.5’s performance to its top reasoning model, deep research, on the SimpleQA test. Deep research’s performance on this benchmark has not been disclosed. Interestingly, Perplexity’s Deep Research model—similar in capability to OpenAI’s deep research—surpassed GPT-4.5 in factual accuracy on this benchmark.
We also recommend
- Secure Your Crypto – The Best Wallet Apps for Maximum Protection
- Best Crypto Apps for Earning Passive Income in 2025
- A 4-Week Strategy to Landing a UK Job with Visa Sponsorship Offers.
- Bybit Hack: North Korean Cybercriminals Strike Again, Stealing $1.5 Billion in Crypto
