GPT-4o

GPT-4o ("o" for "omni") is a multilingual, multimodal generative pre-trained transformer developed by OpenAI and released in May 2024.^[1] GPT-4o is free, but with a usage limit that is five times higher for ChatGPT Plus subscribers.^[2] It can process and generate text, images and audio.^[3] Its application programming interface (API) is twice as fast and half the price of its predecessor, GPT-4 Turbo.^[1]

Originally, multiple versions of GPT-4o were secretly launched under different names on Large Model Systems Organization's (LMSYS) Chatbot Arena as three different models. These three models were called gpt2-chatbot, im-a-good-gpt2-chatbot, and im-also-a-good-gpt2-chatbot.^[4] On 7 May 2024, Sam Altman tweeted "im-a-good-gpt2-chatbot", which was commonly interpreted as a confirmation that these were new OpenAI models being A/B tested.^[5]

GPT-4o achieved state-of-the-art results in voice, multilingual, and vision benchmarks, setting new records in audio speech recognition and translation.^[6][7] GPT-4o scored 88.7 on the Massive Multitask Language Understanding (MMLU) benchmark compared to 86.5 by GPT-4.^[8] Unlike GPT-3.5 and GPT-4, which rely on other models to process sound, GPT-4o natively supports voice-to-voice.^[8] Sam Altman noted on 15 May 2024 that GPT-4o's voice-to-voice capabilities were not yet integrated into ChatGPT, and that the old version was still being used.^[9] This new mode, called Advanced Voice Mode, is currently in limited alpha release.^[10]

The model supports over 50 languages,^[1] which OpenAI claims cover over 97% of speakers.^[11] Mira Murati demonstrated the model's multilingual capability by speaking Italian to the model and having it translate between English and Italian during the live-streamed OpenAI demonstration event on 13 May 2024. In addition, the new tokenizer uses fewer tokens for certain languages, especially languages that are not based on the Latin alphabet, making it cheaper for those languages.^[8]

GPT-4o has knowledge up to October 2023,^[12][13] but can access the Internet if up-to-date information is needed. It has a context length of 128k tokens^[12] with output token limit capped to 4,096,^[13] and after a later update (gpt-4o-2024-08-06) to 16,384.^[14]

As of May 2024, it is the leading model in the LMSYS Elo Arena Benchmarks by the University of California, Berkeley.^[15]

In August 2024, OpenAI introduced a new feature allowing corporate customers to customize GPT-4o using proprietary company data. This customization, known as fine-tuning, enables businesses to adapt GPT-4o to specific tasks or industries, enhancing its utility in areas like customer service and specialized knowledge domains. Previously, fine-tuning was available only on the less powerful model GPT-4o mini.^[16][17]

The fine-tuning process requires customers to upload their data to OpenAI's servers, with the training typically taking one to two hours. Initially, the customization will be limited to text-based data. OpenAI's focus with this rollout is to reduce the complexity and effort required for businesses to tailor AI solutions to their needs, potentially increasing the adoption and effectiveness of AI in corporate environments.^[18][16]

On July 18, 2024, OpenAI released a smaller and cheaper version, GPT-4o mini.^[19]

According to OpenAI, its low cost is expected to be particularly useful for companies, startups and developers that seek to integrate it into their services, which often make a high amount of API calls. Its API costs $0.15 per million input tokens and $0.6 per million output tokens, compared to $5 and $15 respectively for GPT-4o. It is also significantly more capable and 60% cheaper than GPT-3.5 Turbo, which it replaced on the ChatGPT interface.^[19] The price after fine-tuning doubles: $0.3 per million input tokens and $1.2 per million output tokens.^[20]

GPT-4o mini is the default model for users not logged in and use ChatGPT as guests and for those who have hit the limit for GPT-4o.

GPT-4o mini will become available in fall 2024 on Apple's mobile devices and Mac desktops, through the Apple Intelligence feature.^[19]

As released, GPT-4o offered five voices: Breeze, Cove, Ember, Juniper and Sky. A similarity between the voice of American actress Scarlett Johansson and Sky was quickly noticed. On May 14, Entertainment Weekly asked themselves whether this likeness was on purpose.^[21] On May 18, Johansson's husband, Colin Jost, joked about the similarity in a segment on Saturday Night Live.^[22] On May 20, 2024, OpenAI disabled the Sky voice, issuing a statement saying "We've heard questions about how we chose the voices in ChatGPT, especially Sky. We are working to pause the use of Sky while we address them."^[23]

Scarlett Johansson starred in the sci-fi movie Her in 2013, playing the role of Samantha, an artificially intelligent virtual assistant personified through a female voice. As part of the promotion leading up to the release of GPT-4o, Sam Altman on May 13 tweeted a single word: "her".^[24][25]

OpenAI claimed that each voice was based on the voice work of a hired actor. Specifically, OpenAI claimed "Sky’s voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice."^[23] Already in September 2023, OpenAI had claimed their upcoming new talking version of its ChatGPT assistant that sounded like Scarlett Johansson "wasn't meant to resemble" the actress.^[26] CTO Mira Murati stated "I don't know about the voice. I actually had to go and listen to Scarlett Johansson's voice." OpenAI further claimed the voice talent was recruited before reaching out to Johansson.^[25]

On May 21, Johansson issued a statement explaining that OpenAI had repeatedly offered to make her a deal to gain permission to use her voice as early as nine months prior to release, a deal she rejected. She said she was "shocked, angered and in disbelief that Mr. Altman would pursue a voice that sounded so eerily similar to mine that my closest friends and news outlets could not tell the difference." In the statement, Johansson also used the incident to draw attention to the lack of legal safeguards around the use of creative work to power leading AI tools, as her legal counsel demanded OpenAI detail the specifics of how the Sky voice was created.^[25][27]

Observers noted similarities to how Johansson had previously sued and settled with The Walt Disney Company for breach of contract over the direct-to-streaming rollout of her Marvel film Black Widow,^[28] a settlement widely speculated to have netted her around $40M.^[29]

Also on May 21, Shira Ovide at The Washington Post shared her list of "most bone-headed self-owns" by technology companies, with the decision to go ahead with a Johansson sound-alike voice despite her opposition and then denying the similarities ranking 6th.^[26] On May 24, Derek Robertson at Politico wrote about the "massive backlash", concluding that "appropriating the voice of one of the world's most famous movie stars — in reference [...] to a film that serves as a cautionary tale about over-reliance on AI — is unlikely to help shift the public back into [Sam Altman's] corner anytime soon."^[30]

• Llama (language model)
• Apple Intelligence