One of ChatGPT’s greatest assets is its capacity to help with creative writing. OpenAI reported Wednesday that ChatGPT’s next large language model, GPT-4o, has gained a little performance bump. Users should expect “more natural, engaging, and tailored writing to improve relevance & readability” in the future.
GPT-4o, not to be confused with 01 (previously Project Strawberry), is OpenAI’s most recent publicly available model, outperforming GPT-4 and GPT -3.5. GPT-4o was initially introduced in May 2024, and it provides customers with twice the performance and half the resource cost of its direct predecessor, GPT-4-Turbo, as well as cutting-edge benchmark results in speech, language, and vision activities.
Not only is it more efficient than previous versions, but it also has some new features. The model’s quick response time makes it ideal for real-time translation and chat applications.
For example, ChatGPT’s Advanced Voice Mode would not be conceivable without GPT-4o’s near-instantaneous text, voice, and audio inference (which prior models did not provide). GPT-4o outperforms its predecessors in terms of reasoning skills, as it can grasp a user’s spoken tone and intention, recognize unique features of their tone, pace, and mood, and respond appropriately.
GPT-4o is technically available to all OpenAI subscribers, even the free tier, however, it cannot be used indefinitely. Free-tier users will only be able to access the new model a few times on ChatGPT before being redirected to engage with the smaller GPT-4o-mini model. Plus, Teams and Enterprise subscribers have a rate cap that is nearly five times greater.
GPT-4o-mini is based on the same training data as its larger sister but uses fewer variables (and thus fewer computational resources) in its inference procedures. GPT-40-mini’s lower weight and increased responsiveness have made it suitable for a range of small-scale applications, including computer code development.
According to CNBC, OpenAI claims the GPT-4o mini is “the most capable and cost-efficient small model available today,” outperforming competitors like Google’s Gemini 1.5 Flash, Meta’s Llama 3 8b, and Anthropic’s Claude 3 Haiku in the range of benchmarks.
According to statistics from Artificial Analysis, 40-mini earned 82% on the MMLU reasoning benchmark, outperforming Gemini by 3% and Claude 3 Haiku by 7%.