During its “Spring Update,” OpenAI unveiled significant enhancements to ChatGPT and GPT-4. The introduction of GPT-4o (“Omnimodel”) now allows non-paying users access to a substantially more advanced model than before.
OpenAI had already eliminated the need for logging in, but aims to further reduce access friction for ChatGPT. Following the mobile apps, a macOS application is now available, though an internet connection is still required as ChatGPT cannot run locally.
Additionally, ChatGPT now supports 50 languages, covering 97 percent of the world’s population according to OpenAI. Microsoft Copilot, based on OpenAI technology, also recently added support for more languages.
Enhanced Chat Interface and Real-Time Responses
The interface has been redesigned to resemble a standard chat app more closely. The speech feature has been improved to respond in near real-time, creating a more natural conversational experience. Unlike the previous voice feature, there is no delay after sending audio; ChatGPT can even be interrupted and respond immediately to new statements. Moreover, the chatbot can now watch live video feeds and react in real-time. Similar to Amazon Alexa, ChatGPT activates with the phrase “Hey ChatGPT.”
Key Improvements: Efficiency and Multimodality
The core of the ChatGPT update is increased efficiency. GPT-4o promises similar results to GPT-4 but is 50 percent more efficient, making it significantly cheaper for API use. Previously, OpenAI used multiple models for processing text, video, and speech, which required translation between models and caused latency. Now, everything runs on GPT-4o, making the AI truly “multimodal” and not just text-focused.
During the presentation, OpenAI demonstrated the broader applicability of ChatGPT, including its ability to watch live videos, function as an interpreter, and adjust its intonation as needed.
Free Access to GPTs
Other paid features of ChatGPT Plus are now available for free. Everyone can now access the GPT Store, offering specialized LLMs for programming, academic purposes, or writing poetry. Note that these are third-party modifications of OpenAI’s GPT models, so error rates may vary and user data might be shared with other parties.
Upcoming Global Rollout
At the time of writing, the update is not yet live. OpenAI promises to roll out the new features globally in the coming weeks. In closing, the AI developer thanked Nvidia, whose CEO Jensen Huang personally handed over the first Blackwell AI chip to OpenAI. According to OpenAI, the demonstration (video) would not have been possible without this hardware.
There is currently no mention of an AI search engine from OpenAI, contrary to previous reports from last week.
