According to reports, despite facing financial challenges, OpenAI, a cutting-edge generative AI company, continues to steadily release new models and feature updates.
Recently, OpenAI quietly launched a new webpage and officially introduced a new member of the GPT-4o series - the GPT-4o Long Output language model. This model is an extension of the iconic GPT-4o released in May, with the most notable change being a significant increase in output capacity: a single output can reach 64,000 tokens, a staggering 16-fold increase compared to the initial 4,000 tokens of GPT-4o.
Tokens, as the basic units of processing in large language models, represent the numerical expressions of concepts, grammatical structures, and alphanumeric combinations. They are cleverly organized behind the scenes of the model based on semantic meaning. Through OpenAI's Tokenizer or the interactive tool provided by machine learning expert Simon Willison, we can intuitively understand how tokens actually work.
The launch of GPT-4o Long Output is a direct response from OpenAI to customers' demand for longer outputs. A company spokesperson revealed to VentureBeat, "We have received feedback from customers who desire longer output contexts. Therefore, we are constantly exploring new methods to better meet these needs."
Currently, the model is in the alpha testing phase and is expected to last for several weeks. During this period, OpenAI will collect data and evaluate the effectiveness of this extended output feature in practical applications.
For applications that require detailed and extensive outputs, such as code editing and writing assistance, GPT-4o Long Output is undoubtedly a powerful tool. Its longer output capacity allows the model to provide more comprehensive and detailed responses, greatly enhancing the user experience.
It is worth noting that the GPT-4o series models have always maintained a maximum context window of 128,000 tokens, and this limitation also applies to GPT-4o Long Output. So how did OpenAI achieve a significant increase in the number of output tokens while keeping the total context window unchanged? This is actually achieved by adjusting the ratio of input to output. In GPT-4o Long Output, users can provide up to 64,000 tokens of input in a single interaction and receive an equal amount of output, thus achieving longer replies.
In addition, the pricing strategy of GPT-4o Long Output is also quite proactive, charging $6 per million input tokens and $18 per million output tokens. Compared to other GPT-4o series models, its pricing is more flexible and relatively affordable, reflecting OpenAI's commitment to popularize powerful AI technology among a wider community of developers.
Currently, experimental access to GPT-4o Long Output is limited to a few trusted partners. OpenAI stated that they will consider expanding access to a broader customer base based on feedback from the alpha testing phase.
