Azure OpenAI pro【【微信】】erent models, grouped by family and capability. A model family typically associates models by their intended task. The following table describes model families currently a【【微信】】I. Not all models are a【【微信】】 currently. Refer to the model capability table in this article for a full breakdown.

Each model family has a series of models that are further distinguished by capability. These capabilities are typically identified by names, and the alphabetical order of these names generally signifies the relati【【微信】】f that model within a given model family. For example, GPT-3 models use names such as Ada, Babbage, Curie, and Da【【微信】】ive capability and cost. Da【【微信】】d more expensive than Curie, which in turn is more capable and more expensi【【微信】】, and so on.

【【微信】】s typically correspond to the following standard naming con【【微信】】:

For example, our most powerful GPT-3 model is called , while our most powerful Codex model is called .

The older 【【微信】】s named , , , and that don't follow the standard naming con【【微信】】 are primarily intended for fine tuning. For more information, see Learn how to customize a model for your application.

You can get a list of models that are a【【微信】】ence and fine-tuning by your Azure OpenAI resource by using the Models List API.

We recommend starting with the most capable model in a model family to confirm whether the model capabilities meet your re【【微信】】. Then you can stay with that model or mo【【微信】】apability and cost, optimizing around that model's capabilities.

GPT-4 can sol【【微信】】ith greater accuracy than any of OpenAI's pre【【微信】】. Like gpt-35-turbo, GPT-4 is optimized for chat but works well for traditional completions tasks.

These models are currently in preview. 【【微信】】, existing Azure OpenAI customers can apply by filling out this form.

The supports 8192 max inpu【【微信】】 and the supports up to 32,768 tokens.

The GPT-3 models can understand and generate natural language. The ser【【微信】】apabilities, each with different le【【微信】】uitable for different tasks. Davinci is the most capable model, while Ada is the fastest. In the order of greater to lesser capability, the models are:

While Davinci is the most capable, the other models pro【【微信】】vantages. Our recommendation is for users to start with Da【【微信】】ing, because it produces the best results and 【【微信】】 Azure OpenAI can provide. Once you ha【【微信】】, you can then optimize your model choice with the best latency/performance balance for your application.

Davinci is the most capable model and can perform any task the other models can perform, often with less instruction. For applications re【【微信】】g of the content, like summarization for a specific audience and creati【【微信】】, Da【【微信】】 results. The increased capabilities pro【【微信】】e more compute resources, so Da【【微信】】't as fast as other models.

Another area where Da【【微信】】standing the intent of text. Da【【微信】】lving many kinds of logic problems and explaining the moti【【微信】】. Davinci has been able to solve some of the most challenging AI problems in【【微信】】t.

Use for: Complex intent, cause and effect, summarization for audience

Curie is powerful, yet fast. While Da【【微信】】t comes to analyzing complicated text, Curie is capable for many nuanced tasks like sentiment classification and summarization. Curie is also good at answering 【【微信】】ng Q&A and as a general ser【【微信】】.

Use for: Language translation, complex classification, text sentiment, summarization

Babbage can perform straightforward tasks like simple classification. It’s also capable when it comes to semantic search, ranking how well documents match up with search 【【微信】】.

Use for: Moderate classification, semantic search classification

Ada is usually the fastest model and can perform tasks like parsing text, address correction and certain kinds of classification tasks that don’t re【【微信】】. Ada’s performance can often be impro【【微信】】ntext.

Use for: Parsing text, simple classification, address correction, keywords

The ChatGPT model (gpt-35-turbo) is a language model designed for con【【微信】】s and the model behaves differently than pre【【微信】】. Pre【【微信】】 were text-in and text-out, meaning they accepted a prompt string and returned a completion to append to the prompt. However, 【【微信】】versation-in and message-out. The model expects a prompt string formatted in a specific chat-like transcript format, and returns a completion that represents a model-written message in the chat.

To learn more about the ChatGPT model and how to interact with the Chat API check out our in-depth how-to.

The Codex models are descendants of our base GPT-3 models that can understand and generate code. Their training data contains both natural language and billions of lines of public code from GitHub.

They’re most capable in Python and proficient in o【【微信】】, including C#, Ja【【微信】】, Go, Perl, PHP, Ruby, Swift, TypeScript, SQL, 【【微信】】. In the order of greater to lesser capability, the Codex models are:

Similar to GPT-3, Davinci is the most capable Codex model and can perform any task the other models can perform, often with less instruction. For applications re【【微信】】g of the content, Da【【微信】】 results. Greater capabilities re【【微信】】 Da【【微信】】't as fast as other models.

Cushman is powerful, yet fast. While Da【【微信】】t comes to analyzing complicated tasks, Cushman is a capable model for many code generation tasks. Cushman typically runs faster and cheaper than Davinci, 【【微信】】.

Currently, we offer three families of Embeddings models for different functionalities:

• Similarity
• Text search
• Code search

Each family includes models across a range of capability. The following list indicates the length of the numerical 【【微信】】ervice, based on model capability:

• Ada: 1024 dimensions
• Babbage: 2048 dimensions
• Curie: 4096 dimensions
• Davinci: 12288 dimensions

Davinci is the most capable, but is slower and more expensi【【微信】】. Ada is the least capable, but is both faster and cheaper.

These models are good at capturing semantic similarity between two or more pieces of text.

These models help measure whether long documents are rele【【微信】】ery. There are two input types supported by this family: , for embedding the documents to be retrieved, and , 【【微信】】h query.

Similar to text search embedding models, there are two input types supported by this family: , for embedding code snippets to be retrieved, and , 【【微信】】anguage search 【【微信】】.

When using our embeddings models, keep in mind their limitations and risks.

These models can be used with Completion API re【【微信】】. is the only model that can be used with both Completion API re【【微信】】letion API.