OpenAI’s GPT-4.5 has arrived.
On Thursday, the ChatGPT maker announced the latest version of its LLM (large language model) in research preview. As OpenAI CEO Sam Altman said a few weeks ago, GPT-4.5 is OpenAI’s last non-chain-of-thought model, meaning all future models will have reasoning capabilities.
“Early testing shows that interacting with GPT-4.5 feels more natural,” said the white paper. “Its broader knowledge base, stronger alignment with user intent, and improved emotional intelligence make it well-suited for tasks like writing, programming, and solving practical problems – with fewer hallucinations.”
In the demo, OpenAI team members talked about how GPT-4.5 is not a reasoning model, but it still has advanced knowledge and can be used to reason through problems, even though it’s meant to be used as general purpose tool compared to the o series which is best for science and mathematics.
GPT-4.5 is the newest addition to its GPT series since OpenAI announced GPT-4o last May. OpenAI has also shipped versions of reasoning models, o1, o3-mini, and o3-mini-high in the past year in a reported effort to focus more on models that rely on increased computing power (test-time compute) in real-time instead of extensive training and more data.
Mashable Light Speed
The release of GPT-4.5 also serves as a reminder that the rollout of the elusive GPT-5 has not gone to plan. Persistent rumors indicate that OpenAI has seen diminishing returns in model improvements, suggesting scaling laws (more compute, bigger datasets) for training a model might not equate to improved results.
This is a possible reason why we’ve yet to see GPT-5 which has been massively hyped. Altman said it will be smarter than him and former CTO Mira Murati who said it will have “Ph.D.-level intelligence.” In the meantime, we now have GPT-4.5, which could either be an indication that OpenAI is close to meeting very high expectations or stalling for more time.
Anonymous former OpenAI employees told Fortune that GPT-4.5, internally dubbed Orion, was supposed to be GPT-5, but couldn’t achieve enough performance gains to merit the GPT-5 name. Altman said GPT-5 could be expected in a matter of months.
How to try GPT-4.5
As of today, we can start putting GPT-4.5 to the test. Well, not everyone. GPT-4.5 is currently only available to ChatGPT Pro members who shell out $200 a month, with access for ChatGPT Plus subscribers who pay $20 a month, coming out next week. It’s only a week apart, but Plus subscribers used to be the top tier granted with early access to new features are now wondering why they even bother paying for a subscription.
While OpenAI is undoubtedly hoping to gain more Pro subscribers, this could have a backfiring effect by alienating Plus subscribers who suddenly have good options with competitors that are catching up, like xAI’s Grok 3, DeepSeek R1, and Anthropic’s Claude 3.7 Sonnet.
A lot is riding on GPT-4.5 as an indicator of GPT-5 progress, and now we’ll know whether it lives up to the hype.