OpenAI announces GPT-4 upgraded large-scale language model ``GPT-4 Turbo'', has knowledge until April 2023, context window is 128K, and price is modest
OpenAI, an AI development company, announced ' GPT-4 Turbo ', an upgraded version of its large-scale language model 'GPT-4', at the annual developer event '
New models and developer products announced at DevDay
https://openai.com/blog/new-models-and-developer-products-announced-at-devday
OpenAI introduces GPT-4 Turbo: Larger memory, lower cost, new knowledge | Ars Technica
In large-scale language models such as GPT-4 Turbo, the amount of text that can be processed at once to generate a response is important, called the ' context window .' The longer the context window, the more likely the large language model will be able to generate long sentences without losing memory about the topic.
Previous GPT-4 context windows were 8,000 tokens, with a 32,000-token model available to some developers. However, the GPT-4 Turbo announced this time has a context window of 128,000 tokens, which is approximately 96,000 words, which is equivalent to 300 pages of a book. It is also said to be more accurate than previous models.
In addition, the previous GPT-4 was trained with information up to September 2021, and gave inaccurate answers to questions regarding the latest information. However, GPT-4 Turbo has knowledge up to April 2023, making it the latest large-scale language model in OpenAI. Furthermore, OpenAI has revealed that it will frequently update information on GPT-4 Turbo.
Furthermore, with GPT-4 Turbo, in `` Function calling '' it is now possible to call multiple actions from a single message, such as ``open the car windows and turn off the air conditioner'', and ``always respond in XML.'' The accuracy has been improved in the generation of certain formats such as ``. It also newly supports ' JSON mode '.
In addition, the 'seed' parameter ensures that the language model returns consistent responses and allows for reproducible output. This feature is useful for debugging and unit testing.
OpenAI also announced the release of a new version of the previous GPT-3.5 Turbo. The new GPT-3.5 Turbo has a 16,000-token context window and supports generation in specific formats, JSON mode, and function calling that can be executed with GPT-4 Turbo. Developers can use the new GPT-3.5 Turbo with model ID 'gpt-3.5-turbo-1106' starting December 11, 2023.
While GPT-4 Turbo has improved performance over GPT-4, the price of GPT-4 Turbo is modest. Below is the price list per token for GPT-4 Turbo and GPT-3.5 Turbo announced this time. The input of GPT-4 Turbo is 1/3 and the output is 1/2 compared to the conventional GPT-4. In addition, the price of the GPT-3.5 Turbo 16K has also been reduced from the previous model.
| input | output | |
| GPT-4 8K | $0.03 (approx. 4.5 yen) | $0.06 (approx. 9.0 yen) |
| GPT-4 32K | $0.06 (approx. 9.0 yen) | $0.012 (approx. 1.8 yen) |
| GPT-4 Turbo 128K | $0.01 (approx. 1.5 yen) | $0.03 (approx. 4.5 yen) |
| GPT-3.5 Turbo 4K | 0.0015 dollars (approx. 0.23 yen) | 0.002 dollars (approximately 0.3 yen) |
| GPT-3.5 Turbo 16K (old) | 0.003 dollars (approximately 0.45 yen) | 0.004 dollars (approximately 0.6 yen) |
| GPT-3.5 Turbo 16K(New) | $0.001 (approx. 0.15 yen) | 0.002 dollars (approximately 0.3 yen) |
Developers can try out GPT-4 Turbo with the model ID 'gpt-4-1106-preview', and OpenAI has revealed that it plans to release a stable production version in the coming weeks.
Related Posts:
in Software, Posted by log1r_ut