OpenAI Launches New GPT-4o Model For Free

OpenAI on Monday unveiled a new flagship AI model that could talk, listen, and see. Touted as a step forward in AI, the new “GPT-4o” is billed as a faster and more accurate version of GPT-4. Unlike the GPT-4 model, which remains available only to paid subscribers, GPT-4o is offered for free.

In videos released by OpenAI, GPT-4o was shown holding conversations with OpenAI research leads and even reading expressions using the phone's camera to judge their emotions. Two instances of GPT-4o on separate smartphones could also hold a conversation and sing along with lyrics describing the room where the recording was done.

According to OpenAI, these are possible due to the fast reaction times of GPT-4o. Averaging just 320 milliseconds, this is similar to human response time in a conversation, OpenAI says. GPT-4o is also faster than GPT-4 while matching the performance of the best GPT-4 Turbo model in areas such as English and code generation.

“It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models,” wrote OpenAI in a blog post.

Available for free

Performance aside, the more astounding aspect of GPT-4o is how it is available for free. GPT-4o will be offered to ChatGPT Free users with “usage limits,” wrote OpenAI in a separate announcement. Plus users will have a message limit that is “5x greater than free users,” while Team and Enterprise users will have “even higher limits.”

Considering that OpenAI is first to market with an easy-to-use multimodal AI optimized for conversation and vision with GPT-4 capabilities, it is clear that it hopes to gain greater market share in a bid to stay competitive, attract investment, and shape the future of generative AI.

The decision to offer GPT-4o for free is no doubt a result of its lower cost; API access to GPT-4o is priced at half the cost of GPT-4.

Moreover, access to more users also translates to much more data that OpenAI can access through ChatGPT, particularly speech and visual data. While it is possible to disallow the use of such data for training future AI models, it’s not a default configuration and a significant proportion of users will likely omit to do so.

The GPT-4o announcement was made a day ahead of the Google I/O 2024 conference, where Google unveiled its AI vision. Google also announced Gemini 1.5 Flash, a cost-effective model designed to quickly summarize conversations, caption images and videos, and extract data from large documents and tables.

Image credit: OpenAI (Video still)