OpenAI has introduced an updated GPT-4o model with significant improvements. The new version creates more natural and relevant texts and provides deeper analysis of uploaded files.
ChatGPT users can already test the update. Developers have access to the model via APIs called gpt-4o-2024-11-20 and chatgpt-4o-latest. The technical specifications remain unchanged: a context window of 128,000 tokens, a maximum output of —16,384 tokens, and training data until October 2023.
For a week, the model was tested anonymously on the Chatbot Arena platform. Based on more than 8,000 user votes, GPT-4o outperformed Google’s Gemini-Exp-1114, which received 1,361 points.
In the creative writing category, the new version improved its score from 1365 to 1402 points. The GPT-4o also moved up from second to first place in the overall rating, programming, and complex tasks, and from fourth to third place in math.
The updated version of the GPT-4o demonstrates a shift in OpenAI’s priorities away from logical reasoning, which was focused on model o1and creative writing. This sets the company apart from the industry-wide focus on math and programming.
Source: Neowin