OpenAI Announces General Availability of GPT-4 API and Deprecation Plan for Older Models

0
49
GPT-4 API Now Available to All Paying Users: OpenAI Announces Access Expansion
OpenAI Provides GPT-4 API Access to API Users with Successful Payment History

OpenAI Introduces GPT-4 API and Shifts Focus to Chat Completions API for Enhanced Developer Experience

Key Points
  • OpenAI launches the GPT-4 API, fulfilling millions of developer requests for access to the advanced language model.
  • GPT-4 API now available to existing API developers, with plans to open up access to new developers soon.
  • GPT-3.5 Turbo, DALL·E, and Whisper APIs are also generally available, providing developers with additional powerful AI models.
  • OpenAI emphasizes the transition from text completions to chat completions, announcing a deprecation plan for older models of the Completions API.
  • The Chat Completions API, introduced in March, accounts for 97% of API GPT usage, offering enhanced conversational capabilities.
  • Older completion models will be replaced with new models, and developers are recommended to adopt the Chat Completions API.
  • OpenAI provides details on the deprecation timeline and new model names for a smooth transition.
  • Developers using fine-tuned models are advised to prepare for the transition by fine-tuning atop new base GPT-3 models or newer models.

OpenAI, the leading artificial intelligence (AI) research organization, has announced the general availability of the highly popular GPT-4 API, catering to the demands of millions of developers worldwide. The GPT-4 API represents OpenAI’s most capable model to date, with a wide range of innovative products already leveraging its powerful language processing capabilities. As of today, existing API developers with a proven history of successful payments can access the GPT-4 API, allowing them to harness its cutting-edge features with an 8K context. OpenAI plans to gradually expand access to new developers by the end of this month, with further rate-limits adjustments based on compute availability.

In addition to the GPT-4 API launch, OpenAI has made significant progress in making other advanced AI models accessible. The GPT-3.5 Turbo, DALL·E, and Whisper APIs have reached general availability, providing developers with an array of powerful tools for their AI-driven projects. OpenAI is actively working on enabling fine-tuning for GPT-4 and GPT-3.5 Turbo, aiming to offer this feature later this year, further expanding the customization options for developers.

OpenAI’s vision for the future revolves around chat-based models that can support various use cases. The company has observed that the Chat Completions API, introduced in March, now accounts for a staggering 97% of API GPT usage. The chat-based paradigm has proven to be exceptionally powerful, catering to a wide range of conversational needs while offering increased flexibility and specificity. Its structured interface, including system messages and function calling, along with multi-turn conversation capabilities, enables developers to create immersive conversational experiences and tackle diverse completion tasks. The Chat Completions API also enhances security by structurally separating user-provided content from instructions, mitigating the risk of prompt injection attacks. OpenAI is committed to investing in this direction and plans to allocate most of its platform efforts to further improve the Chat Completions API, ensuring developers enjoy an increasingly capable and user-friendly experience. The company is actively addressing remaining areas for improvement, such as log probabilities for completion tokens and enhanced steerability to reduce excessive verbosity in responses.

As part of its commitment to optimizing compute capacity and prioritizing the Chat Completions API, OpenAI has announced a deprecation plan for older models of the Completions API. While the API will remain accessible, OpenAI will label it as “legacy” in the developer documentation starting today. This strategic shift reflects OpenAI’s dedication to focusing on the Chat Completions API for future model and product improvements, indicating no plans for publicly releasing new models using the Completions API.

Effective January 4, 2024, users who are currently utilizing older embeddings models, such as text-search-davinci-doc-001, will be required to transition to text-embedding-ada-002. The release of text-embedding-ada-002 in December 2022 has demonstrated its enhanced capabilities and cost-effectiveness compared to previous models. In fact, text-embedding-ada-002 now accounts for an impressive 99.9% of all embedding API usage.

    OpenAI acknowledges that this change presents a significant adjustment for developers who rely on these older models. The decision to phase out these models was not taken lightly. To ensure a seamless transition, OpenAI is committed to covering the financial costs associated with re-embedding content using the new text-embedding-ada-002 model. In the upcoming days, OpenAI will proactively reach out to impacted users, providing them with the necessary support and guidance during this process.

    Applications utilizing the stable model names for base GPT-3 models, namely ada, babbage, curie, and davinci, will automatically be upgraded to the respective new models listed above on January 4, 2024. The new models will be available for early testing in the coming weeks, allowing developers to familiarize themselves with the updated capabilities by specifying the corresponding model names in their API calls.

    Developers currently using other older completion models, such as text-davinci-003, are required to manually upgrade their integration by January 4, 2024. OpenAI advises developers to specify “gpt-3.5-turbo-instruct” in the “model” parameter of their API requests. This new model, based on the InstructGPT-style, is trained similarly to text-davinci-003 and can be seamlessly integrated as a drop-in replacement in the Completions API. OpenAI will make the gpt-3.5-turbo-instruct model available for early testing in the coming weeks.

    For developers wishing to continue utilizing their fine-tuned models beyond January 4, 2024, OpenAI recommends fine-tuning replacements using the new base GPT-3 models (ada-002, babbage-002, curie-002, davinci-002), or newer models such as gpt-3.5-turbo and gpt-4. To ensure a smooth transition, users who previously fine-tuned older models will be granted priority access to GPT-3.5 Turbo and GPT-4 fine-tuning once this feature becomes available later this year. OpenAI acknowledges the challenges involved in migrating off models fine-tuned on user data and commits to providing support to users during this transition, aiming to make it as seamless as possible.

    In the coming weeks, OpenAI plans to reach out to developers who have recently used these older models, providing them with more information once the new completion models are ready for early testing.

    OpenAI also announced the deprecation of older embeddings models, urging users of models like text-search-davinci-doc-001 to migrate to text-embedding-ada-002 by January 4, 2024. The newer model, text-embedding-ada-002, released in December 2022, has demonstrated superior capabilities and cost-effectiveness, becoming the preferred choice for 99.9% of all embedding API usage. OpenAI acknowledges the significant change for developers using these older models andwill cover the financial cost associated with re-embedding content using the new models. OpenAI will be reaching out to impacted users in the coming days to provide further guidance and support.

    Lastly, OpenAI announced the deprecation of the Edits API and its associated models, including text-davinci-edit-001 and code-davinci-edit-001. Developers currently utilizing these models are advised to migrate to GPT-3.5 Turbo by January 4, 2024. The Edits API beta, designed to enable developers to return edited versions of prompts based on instructions, has served as a valuable learning experience for OpenAI. The feedback received from the Edits API has informed the development of gpt-3.5-turbo and the Chat Completions API, both of which can now be used to achieve similar outcomes.

    OpenAI’s commitment to providing developers with cutting-edge AI capabilities and an enhanced developer experience is evident through the launch of the GPT-4 API and the focus on the Chat Completions API. By deprecating older models and promoting the adoption of newer and more efficient models, OpenAI aims to enable developers to create powerful and engaging conversational experiences while ensuring a seamless transition to the latest AI technologies.

    As OpenAI continues to innovate in the field of natural language processing, developers can look forward to even more advanced AI models and tools in the future, further pushing the boundaries of what is possible with AI-powered applications.

    LEAVE A REPLY

    Please enter your comment!
    Please enter your name here
    Captcha verification failed!
    CAPTCHA user score failed. Please contact us!