During a recent live stream, OpenAI has officially unveiled its latest language model, GPT-4.5. However, there is a catch: Accessing this advanced tool requires a ChatGPT Pro subscription, which costs $200 per month.

This new model is still in what the company refers to as a “research preview,” suggesting that users are effectively beta testers who are paying for this privilege.

The company claims that GPT-4.5 is designed to enable more “conversationally natural” interactions and showcases enhanced performance in areas such as programming, pattern recognition, writing, and practical problem-solving tasks.
Importantly, GPT-4.5 should not be considered a reasoning-focused model. It lacks the self-correcting and in-depth research capabilities seen in other models like OpenAI o1, o3-mini, and the anticipated full version of o3. According to a recently leaked document, OpenAI does not classify GPT-4.5 as a “frontier” model, but it is indeed the largest LLM they’ve developed to date. This size is expected to foster quicker responses to everyday inquiries by leveraging its extensive corpus of training data.

Credit: OpenAI

In preliminary evaluations, OpenAI reported that the new model exhibited 24.7% fewer instances of “hallucination” compared to GPT-4.0, while also presenting 34.3% greater accuracy. Notably, around 57% of internal evaluators expressed a preference for GPT-4.5 over its predecessor.

Credit: OpenAI

In essence, GPT-4.5 may not represent the forefront of AI advancements; however, it does offer improvements over existing non-reasoning models. It positions itself as a reasonable alternative between lower-cost, traditional language models and their often premium counterparts. Features such as multimodal capabilities—like AI Voice Mode and video input—are notably absent. Nevertheless, OpenAI is reportedly working on enhancing the user experience, suggesting that some of these features may eventually be integrated into the model.

This development is consistent with the status of the release, as OpenAI emphasizes it is actively investigating the capabilities of GPT-4.5 and its potential applications. In line with this, plans are underway for this model to be rolled out to all paid ChatGPT subscriptions over the next fortnight, before likely becoming available to free users as a substitute for GPT-4.0 when it transitions out of its preview phase.