Posted by Mathieu Tancrez on 5/13/2024
OpenAI has just released its new model, GPT-4o, which is available to non-paying users through its chatbot ChatGPT and via the API at half the cost of its previous model, GPT-4 Turbo. The “o” in GPT-4o stands for “omni,” reflecting the model's enhanced versatility in handling various input and output modalities. This new model boasts improved capabilities in vision and audio understanding.
During the livestream event for the release, CTO Mira Murati asked the chatbot to tell a bedtime story about robots and love, which GPT-4o performed with minimal latency. This marks a significant breakthrough in large language models, making interactions as smooth as conversing with a human and enabling many new use cases. The new model can translate conversations between people speaking different languages in real time.
Compared to other GPT models, GPT-4o is faster and costs half as much as GPT-4 Turbo when accessed via the API. In terms of throughput and latency, it competes with top models like Llama3, Claude 3 Haiku, and Mixtral 8x7B.
Here is a comparison of the latest GPT models:
| GPT-4o (2024-05-13) |
GPT4-Turbo (turbo-2024-04-09) |
GPT4 (0613) |
|
|---|---|---|---|
| Price | $5 / 1 million tokens for inputs $15 / 1 million tokens for outputs |
$10 / 1 million tokens for inputs $30 / 1 million tokens for outputs |
$30 / 1 million tokens for inputs $60 / 1 million tokens for outputs |
| Knowledge cutoff date | October 2023 | December 2023 | September 2021 |
| Context window | 128,000 | 128,000 | 8,192 |