At its inaugural developer conference, OpenAI made waves by introducing GPT-4 Turbo, an upgraded version of its renowned text-generating AI model, GPT-4. What sets this new model apart is its claim to be both “more powerful” and more cost-effective.
GPT-4 Turbo comes in two flavours: one dedicated to text analysis and another that possesses the ability to comprehend both text and images within context. The text-analysing model is currently available in preview through an API, with plans for both versions to be generally accessible in the near future.
The pricing structure for GPT-4 Turbo is set at $0.01 per 1,000 input tokens (roughly equivalent to 750 words) and $0.03 per 1,000 output tokens. For the image-processing GPT-4 Turbo, the pricing depends on the size of the image, with costs starting at $0.00765 for an image with 1080×1080 pixels. OpenAI has managed to reduce the price, offering GPT-4 Turbo at a rate that is 3 times more affordable for input tokens and 2 times more budget-friendly for output tokens compared to GPT-4.
GPT-4 Turbo boasts several enhancements over its predecessor, including a more recent knowledge base to draw from when providing responses. Similar to other language models, GPT-4 Turbo functions as a statistical tool, predicting words based on patterns learned from an extensive dataset. The difference is that GPT-4 Turbo’s knowledge cutoff extends to April 2023, offering more precise responses to inquiries about recent events.
Another significant improvement is the expanded context window of GPT-4 Turbo, allowing it to consider 128,000 tokens, four times the size of GPT-4’s context window and the most extensive of any commercially available model.
Moreover, GPT-4 Turbo introduces a “JSON mode” to ensure valid JSON responses, making it valuable for web applications that require data interchange. Additional parameters provide options to enhance the model’s ability to provide consistent completions and log probabilities for specific applications.
But OpenAI hasn’t forgotten GPT-4 in its pursuit of excellence. An experimental access program for fine-tuning GPT-4 is being launched, although it comes with more oversight and guidance from OpenAI teams due to technical challenges. OpenAI has also doubled the tokens-per-minute rate limit for paying GPT-4 customers, while pricing remains consistent.
These innovations mark a significant leap in AI capabilities, making it more accessible and powerful, which can potentially revolutionize various industries.