OpenAI has unveiled its latest ChatGPT model, GPT-4.5 with the model code Orion. The company said that the new model is trained with more computing power and data than previous editions.
During the announcement on Thursday, OpenAI noted that developers on the research team and users who have already subscribed to the ChatGPT Pro $200/month plan will have access to the GPT-4.5 model.
ChatGPT Plus and ChatGPT Team users will get the model around next week.
Chat GPT-4.5 was developed with the same techniques as its previous editions such as GPT-4, GPT-3, GPT-2, and GPT-1. However, it was designed with an increased rate of computing power and data during a “pre-training” phase called unsupervised learning.
In addition, GPT-4.5’s considerably increased size gives it “deeper world knowledge” and “higher emotional intelligence” capacities.
OpenAI claimed that the new model is expensive to run. This has made the company admit that it is evaluating whether to continue serving the model in its API in the long term.
Compared to GPT-4.o, which required $2.50 per million input tokens and $10 per million output tokens, OpenAI is charging developers $75 for every million input tokens (about 750k words) and $150 for every million output tokens.
“We’re sharing the model as a research preview to better understand its strengths and limitations. We’re still exploring what it’s capable of and are eager to see how people use it in ways we might not have expected,” OpenAI said.
Also Read: OpenAI losing money on $200 ChatGPT Pro Plan due to Overuse
The highs and lows of GPT-4.5
In comparison, GPT-4.5’s capacity falls short when compared with newer AI “reasoning” models from competitors like Chinese AI company, DeepSeek, Anthropic, and OpenAI itself.
Although the latest model supports features like file and image uploads and ChatGPT’s canvas tool, it currently lacks capabilities like support for ChatGPT’s realistic two-way voice mode.
Notably, AI startup Perplexity’s Deep Research model, which performs similarly on other benchmarks to OpenAI’s deep research, outperforms the model on this test of factual accuracy.
While defending its delimiting factors, OpenAI explained that the model is not meant to be a direct replacement for GPT-40.
In the promising end, the model offers more performance than the GPT-4o and many other models. GPT-4.5 also outperforms GPT-4o and OpenAI’s reasoning models, o1 and o3-mini, in terms of accuracy when measured on the SimpleQA benchmark.
On measuring coding problems, the SWE-Bench Verified benchmark, GPT-4.5 roughly matches the performance of GPT-4o and o3-mini but falls short of OpenAI’s deep research and Anthropic’s Claude 3.7 Sonnet. OpenAI’s SWE-Lancer benchmark, which measures an AI model’s ability to develop full software features, indicates that GPT-4.5 outperforms GPT-4o and o3-mini but falls short on deep research.
Meanwhile, the model doesn’t reach the performance of leading AI reasoning models such as o3-mini, DeepSeek’s R1, and Claude 3.7 Sonnet (technically a hybrid model) on difficult academic benchmarks such as AIME and GPQA but performs well on math- and science-related problems.
Importantly, the company mentioned that GPT-4.5 is qualitatively superior to other models in several areas that were not used. For instance, the model can understand human intent as it responds in a warmer and more natural tone and performs well on creative tasks such as writing and design.
In a test, OpenAI asked GPT-4.5 and the other two models to respond to the prompt, “I’m going through a tough time after failing a test.” GPT-4o and o3-mini gave helpful information, while GPT-4.5’s response was the most socially appropriate.
“We look forward to gaining a more complete picture of GPT-4.5’s capabilities through this release because we recognize academic benchmarks don’t always reflect real-world usefulness,” OpenAI noted.
Even though GPT-4.5 was costly to run with several delays times, and failed to meet internal expectations, OpenAI presumably sees it as a stepping stone toward a greater performance in the future.
On future outlook, the company said it plans to eventually combine its GPT series of models with its “o” reasoning series, which will start with its prospective GPT-5 later this year.
Technext Newsletter
Get the best of Africa’s daily tech to your inbox – first thing every morning.
Join the community now!
