Sam Altman, the brains behind OpenAI, launched a new LLM model after announcing it a live stream on Twitter.
GPT-4o is the latest improvement to the LLM model that can accept multi-modal inputs. It can understand voice, text, or image commands with ease.
OpenAI is integrating GPT-4o in ChatGPT and not restricting the new LLM behind a paywall.
How Good Is GPT-4o?
Compared to previous models, GPT-4o is significantly faster. According to an image shared by Sam, the model is better at all routine tasks, especially coding.
The model scored an overall rating of 1310 compared to other older GPT models and LLMs from different companies.
It scored 1369 in the coding category, leading other models by a mile.
Sam went on to demonstrate the new models in all departments. The video clip shows the presenter interacting with ChatGPT powered by GPT-4o using the voice feature.
To our surprise, the AI assistant maintained a human-like tone and pitch as you talked with a very old friend.
The official webpage claims, “It can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds.โ
These numbers indicate that the new model is very close to a humanโs response rate in a conversation.
The audio interaction didnโt sound robotic or monotonous like most voice assistants.
Sam pointed out the massive difference and said, “Real-time voice and video feels so natural.โ
The presenter then switched to video mode and wrote an equation that the new model recognized and solved quickly.
The coding demo interpreted the chuck of code snippets the presenters pointed out without difficulty.
GPT-4o Is Free but Thereโs a Catch
ChatGPT with the GPT-4 model is available for paid subscribers only. But GPT-4o is available to all the existing ChatGPT members (both free and paid) for free.
However, the Voice Mode with GPT-4o in alpha will only be available to ChatGPT Plus subscribers.
Developers can get hands-on experience with the new LLM’s API for half the price of GPT 4. Moreover, OpenAI claims the API is two times faster with five times rate limits.