OpenAI appears to be preparing for a preview version of GPT-4.5. The development was spotted within ChatGPT’s Android app as an announced experimental option, although it’s not yet accessible. According to the notification, LLM will initially be available exclusively to Pro subscribers.

GPT-4.5 will likely be the successor to GPT-4o in the free version of ChatGPT, with higher call limits and/or early access to Plus and Pro users. Currently, however, GPT-4.5 is only known through notifications in mobile apps. LLM is not available yet; pressing the notification will have no effect.

This is not the first time the existence of GPT-4.5 has been revealed. At the end of 2023, ChatGPT heard the name of this model in a conversation. Other AI models sometimes misidentify themselves, such as DeepSeek at one point thinking it was ChatGPT.

For professional users only

A new notification invites Pro users to test OpenAI’s latest and greatest models. The Pro subscription costs $200 per month and has fewer restrictions than the $20 Plus subscription or the free plan.

In addition to launching GPT-4.5, OpenAI is also working on improving its web interface. AI researcher Tibor Blaho discovered that functions such as Deep Research, Canvas, and uploading images have been moved. This is part of a redesigned interface that makes it easier for users to switch between models and features.

For in-depth research mode, a new input option has been added. Users can add relevant sources, areas of interest, and documents in order to obtain detailed reports on them. OpenAI has yet to announce when GPT-4.5 and interface enhancements will actually be available.

Competitors are everywhere

OpenAI has recently released o3-mini, an AI model capable of inference while keeping its token payout relatively low. It's much cheaper than the previous O1, but performs close to the O1 in benchmark tests. Still, OpenAI has enough competitors in this space, including far cheaper alternatives. For example, Google has launched the GeminiFlashThinking model. Previously, DeepSeekR1 caused a stir in the AI ​​company stock market because it has similar features to OpenAI's o1, but is based on a model (DeepSeekV3) and is said to cost only $5 million to train. This has made many people suddenly suspicious of the huge capital expenditures planned by AI companies.

Anthropic also released the highly anticipated Claude update 3.7Sonnet this week. In addition, the CLI tool ClaudeCode was also unveiled.