GPT-4 is an expansive multimodal model (capable of processing both text and image inputs to generate text responses), surpassing previous models in solving complex problems with enhanced accuracy due to its extensive general knowledge and sophisticated reasoning skills. This model is accessible through the OpenAI API for subscribers. Similar to gpt-3.5-turbo, GPT-4 is fine-tuned for conversational interactions but also excels in standard completion tasks via the Chat Completions API. Instructions for utilizing GPT-4 are available in text generation guide.
Model Overview
- gpt-4-0125-preview: The newest GPT-4 Turbo model designed to minimize instances of incomplete tasks. More information available. Context Window: 128,000 tokens. Training Data: Until Apr 2023.
- gpt-4-turbo-preview: Currently aligned with gpt-4-0125-preview. Context Window: 128,000 tokens. Training Data: Until Apr 2023.
- gpt-4-1106-preview: An advanced GPT-4 Turbo model with enhanced instruction adherence, JSON mode, consistent outputs, simultaneous function execution, and more. Maximum of 4,096 output tokens. This preview model is not yet recommended for live applications. More details available. Context Window: 128,000 tokens. Training Data: Until Apr 2023.
- gpt-4-vision-preview: GPT-4 version capable of image comprehension, alongside other GPT-4 Turbo features. Maximum of 4,096 output tokens. This preview model version is not yet ready for live use. Further information available. Context Window: 128,000 tokens. Training Data: Until Apr 2023.
- gpt-4: Currently directed to gpt-4-0613. Continuous model upgrades noted. Context Window: 8,192 tokens. Training Data: Until Sep 2021.
- gpt-4-0613: A snapshot of gpt-4 from June 13th, 2023, with enhanced function call support. Context Window: 8,192 tokens. Training Data: Until Sep 2021.
- gpt-4-32k: Currently aligned with gpt-4-32k-0613. This model was not widely released, giving way to GPT-4 Turbo. Context Window: 32,768 tokens. Training Data: Until Sep 2021.
- gpt-4-32k-0613: A snapshot of gpt-4-32k from June 13th, 2023, with improved function call support. This model was not broadly deployed in favor of GPT-4 Turbo. Context Window: 32,768 tokens. Training Data: Until Sep 2021.
For basic tasks, the performance gap between GPT-4 and GPT-3.5 models is minimal. However, GPT-4 significantly surpasses previous models in complex reasoning scenarios.
Multilingual Capabilities
GPT-4 outshines both earlier large language models and, as of 2023, most leading-edge systems (which often include benchmark-specific training or manual tweaking). On the MMLU benchmark, a set of English-language multiple-choice questions covering 57 subjects, GPT-4 not only significantly surpasses existing models in English but also exhibits robust performance in various other languages.
Read more about: