OpenAI announces enhanced text generation capabilities with the introduction of GPT-4 and GPT-3.5-turbo’s function calling feature. The models can now generate code to execute programming functions and offer a larger context window for improved performance. While the expanded GPT-3.5-turbo comes at a higher price, OpenAI reduces prices for the original version and popular text embedding models. The company focuses on incremental updates rather than developing new models from scratch, indicating further work before the successor to GPT-4 is trained.
In the face of growing competition in the generative AI field, OpenAI is making significant improvements to its text-generating models while simultaneously lowering prices.
Today, OpenAI unveiled upgraded versions of GPT-3.5-turbo and GPT-4, the latter being their latest text-generating AI. GPT-4 introduces a remarkable capability called function call, which enables developers to describe programming functions to the models. In turn, the models can generate code to execute those functions. This feature proves useful for creating chatbots that rely on external tools to answer questions, converting natural language into database queries, and extracting structured data from text. OpenAI explains that these models have been fine-tuned to detect when a function needs to be called and respond with JSON that adheres to the function signature. Consequently, developers can rely on function calling to obtain structured data more reliably from the models.
In addition to function calls, OpenAI is introducing an enhanced version of GPT-3.5-turbo with a significantly expanded context window. The context window refers to the amount of text the model considers before generating additional text. Models with limited context windows often forget recent conversations, causing them to veer off-topic, sometimes in problematic ways. The new GPT-3.5-turbo offers a four-fold increase in context length, allowing it to process up to 16,000 tokens. However, this version comes at twice the price, with rates set at $0.003 per 1,000 input tokens and $0.004 per 1,000 output tokens. Although it falls short of AI startup Anthropic’s flagship model, which can handle hundreds of pages, OpenAI’s model can process around 20 pages of text in a single run. Notably, OpenAI is currently testing a limited release of GPT-4 with a 32,000-token context window.
On the pricing front, OpenAI is reducing the costs of GPT-3.5-turbo, the original version without the expanded context window, by 25%. Developers can now utilize this model at $0.0015 per 1,000 input tokens and $0.002 per 1,000 output tokens, allowing approximately 700 pages per dollar.
Furthermore, OpenAI is lowering the pricing for text-embedding-ada-002, one of its popular text-embedding models. Text embeddings gauge the relatedness of text strings and are commonly employed in search and recommendation systems. The cost of text-embedding-ada-002 has been reduced by 75%, now priced at $0.0001 per 1,000 tokens. OpenAI attributes this reduction to increased efficiency in their systems, an area of focus that aligns with the company’s substantial investment in research and infrastructure.
Following the release of GPT-4 in early March, OpenAI has indicated that it will focus on incremental updates to existing models rather than developing entirely new models from scratch. CEO Sam Altman reaffirmed this approach during a recent conference hosted by the Economic Times, emphasizing that OpenAI has yet to commence training the successor to GPT-4. The company has considerable work ahead before undertaking such a project.
