Source of the article: AIGC Open Community
On the early morning of January 26th, OpenAI made a major update to the ChatGPT model on its official website, releasing two new large and small text embedding models, the new GPT-4 Turbo model (fixing lazy behavior), a free audit model, and significantly reducing the price of the new GPT-3.5 Turbo model API.
OpenAI will also launch a new API key and a visual management method to help developers observe API usage more easily and intuitively, and provide more detailed usage permissions for API keys.
It is worth mentioning that the new embedding models can provide technical support for knowledge retrieval in ChatGPT, Assistants API, and many other retrieval-enhanced generative development tools.
New Text Embedding Models
"AIGC Open Community" briefly introduces the embedding model: Embedding is a string of numbers that represents concepts in natural language or code content. Embedding also makes it easier for machine learning models and other algorithms to understand the relationship between content and perform tasks such as classification, content retrieval, search, and recommendation.
At the same time, embedding is a core component of the GPT series models, used to convert input text (words or characters) into numerical vectors, such as word embedding, position embedding, and context embedding. These vectors can represent rich information of the input data, providing a deeper semantic understanding.
This time, OpenAI released the small text embedding model text-embedding-3-small and the large text embedding model text-embedding-3-large, with the following main performance features.
1) Stronger Performance: According to the performance tests published by OpenAI, the average score of the small text embedding model for multilingual retrieval (MIRACL) common benchmark increased from 31.4% to 44.0%; for English tasks (MTEB), the average score increased from 61.0% to 62.3%.
For the large text embedding model, the average score on MIRACL increased from 31.4% to 54.9%, and on MTEB, the average score increased from 61.0% to 64.6%, showing overall stronger performance than the small text embedding model.
2) Support for Shortened Embeddings to Save Costs: Compared to smaller embeddings, developers typically spend more costs using larger embeddings (e.g., storing them in vector storage for retrieval), consuming more AI computing power, memory, and storage space.
To help developers save costs, OpenAI allows developers to shorten the embedding model (remove some numbers from the end of the sequence) by passing the dimension API parameter, without losing its representational properties.
For example, on the MTEB benchmark, the large text embedding model can be shortened to a size of 256, yet its performance is still better than the unshortened small text embedding model with a size of 1536.
3) API Prices: Although the new text embedding models have strong performance, OpenAI has significantly reduced the prices of the APIs. The API price for the small text embedding model has decreased by 5 times compared to the previous model, with a price of 0.00002 USD per 1000 tokens. For the large text embedding model, the price is 0.00013 USD per 1000 tokens.
New GPT-4 Turbo Preview Model
Since OpenAI released the GPT-4 Turbo model, over 70% of GPT-4 API customers have switched to GPT-4 Turbo. This is because GPT-4 Turbo can provide a larger context and better performance.
Now, OpenAI has released the all-new GPT-4 Turbo preview model—gpt-4-0125-preview.
Compared to the previous version, this model can better handle tasks such as code generation, while also fixing the highly anticipated lazy behavior and addressing errors affecting non-English UTF-8 generation.
For developers who wish to automatically upgrade to the latest GPT-4 Turbo preview model, it will always point to OpenAI's latest GPT-4 Turbo preview version.
Free Audit Model
To help developers reduce illegal content output in ChatGPT and improve security, OpenAI provides a free audit model API.
In addition, OpenAI will also release the most powerful audit model to date, text-moderation-007, to further enhance the model's security.
Significant API Price Reduction
Next week, OpenAI will launch the all-new GPT-3.5 Turbo series model—gpt-3.5-turbo-0125, and significantly reduce the API prices.
The input price of the new model has been reduced by 50%, with a price of 0.0005 USD per 1000 tokens; the output price has been reduced by 25%, with a price of 0.0015 USD per 1000 tokens.
At the same time, the model has undergone various functional improvements, including improving the accuracy of responding to requested formats and fixing issues causing text encoding errors in non-English language function calls.
All-New Visual API Management Methods
To help developers manage APIs more efficiently, OpenAI has provided two new management methods.
1) Developers can now assign detailed permissions for API keys from the API key page. For example, they can assign read-only access to support internal dashboards, or restrict access to specific endpoints.
2) After enabling the tracking feature, the usage details and export functions can now display metrics at the API key level. Therefore, developers can easily view detailed usage at the level of each feature, team, product, or project by setting separate API keys for each.
In the coming months, OpenAI will further enhance developers' ability to use, observe, and control APIs, which is crucial for large enterprises.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
