GPT-4 API Released, Marking the Deprecation of Older Models in Completions API

In a significant development for the AI community, OpenAI has announced the general availability of the GPT-4 API, catering to the demands of millions of developers who have eagerly anticipated its release. OpenAI’s GPT-4, touted as their most capable model yet, has already garnered immense interest, with numerous innovative products already leveraging its capabilities.

The GPT-4 API, which was made accessible to all existing API developers with a successful payment history, allows users to tap into the power of this cutting-edge language model. Presently, access to the GPT-4 API is limited to a context size of 8K, but OpenAI plans to expand access to new developers by the end of this month. Moreover, the company intends to gradually raise rate-limits, depending on compute availability, to accommodate the growing demand.

However, GPT-4 is not the only model making headlines today. OpenAI has also announced the general availability of the GPT-3.5 Turbo, DALL·E, and Whisper APIs. These models, known for their stability and readiness for production-scale use, are expected to empower developers with enhanced capabilities and versatility.

One of the noteworthy developments from OpenAI is their increased emphasis on chat-based models, particularly highlighted by the introduction of the Chat Completions API. OpenAI’s observations have revealed that a structured prompt interface can often yield superior results compared to freeform text prompts. Consequently, the chat-based paradigm has gained prominence due to its remarkable ability to handle various use cases and facilitate new conversational needs. This approach provides developers with greater flexibility, specificity, and enables them to create immersive conversational experiences. Moreover, the structured interface, including system messages and function calling, combined with the multi-turn conversation capabilities, enhances the developers’ ability to build interactive and engaging applications. Additionally, this structured approach mitigates the risk of prompt injection attacks by effectively separating user-provided content from instructions.

OpenAI’s commitment to advancing chat completions is reflected in their future plans, as they intend to invest heavily in this direction. Recognizing the potential of chat-based models to offer increasingly capable and user-friendly experiences, OpenAI is dedicated to closing any remaining gaps in the Chat Completions API. They are actively working on introducing features such as log probabilities for completion tokens and increased steerability to reduce the excessive “chattiness” often associated with responses.

In line with their strategy to optimize compute capacity and prioritize chat completions, OpenAI has unveiled a deprecation plan for older models in the Completions API. Although the Completions API will remain accessible, it will be labeled as “legacy” in OpenAI’s developer documentation. The retirement of these older models will commence on January 4, 2024. This move aims to pave the way for future improvements by focusing on the Chat Completions API, signaling that OpenAI does not have plans to release new models based on the Completions API.

To ensure a smooth transition, OpenAI has provided a comprehensive list of older completion models and their corresponding replacements. For instance, the older base GPT-3 models such as ada, babbage, curie, and davinci will be automatically upgraded to their respective new models—ada-002, babbage-002, curie-002, and davinci-002—on January 4, 2024. Additionally, early testing opportunities for these new models will be available in the coming weeks by specifying the corresponding model names in API calls.

Developers currently utilizing other older completion models, such as text-davinci-003, will need to manually upgrade their integration by January 4, 2024, by specifying gpt-3.5-turbo-instruct in the “model” parameter of their API requests. This new model, akin to text-davinci-003, acts as a seamless drop-in replacement within the Completions API. OpenAI plans to release the new model for early testing in the near future.

For developers interested in continuing to use their fine-tuned models beyond January 4, 2024, OpenAI advises fine-tuning replacements based on the new base GPT-3 models (ada-002, babbage-002, curie-002, davinci-002) or newer models (gpt-3.5-turbo, gpt-4). Users with previously fine-tuned older models will be given priority access to GPT-3.5 Turbo and GPT-4 fine-tuning once this feature becomes available later in the year. Recognizing the challenges involved in migrating away from fine-tuned models based on user data, OpenAI is committed to providing support to ensure a seamless transition for users.

Furthermore, OpenAI has also revealed plans to deprecate older embeddings models. Users relying on older embeddings models, such as text-search-davinci-doc-001, will need to migrate to text-embedding-ada-002 by January 4, 2024. OpenAI released text-embedding-ada-002 in December 2022, and it has proven to be more capable and cost-effective than its predecessors. To alleviate the financial burden of this transition, OpenAI has generously offered to cover the cost of re-embedding content with the new models, and they will reach out to impacted users in the coming days to provide more information.

Another notable deprecation pertains to the Edits API and its associated models, such as text-davinci-edit-001 or code-davinci-edit-001. Developers utilizing these models will need to migrate to GPT-3.5 Turbo by January 4, 2024. Initially introduced as an exploratory API, the Edits API aimed to enable developers to receive an edited version of the prompt based on instructions. OpenAI has taken user feedback into account and incorporated it into the development of gpt-3.5-turbo and the Chat Completions API. These alternative options can now be employed for similar purposes and provide a more comprehensive solution.

OpenAI’s latest announcements, including the general availability of the GPT-4 API and the deprecation plans for older models, mark a significant milestone in the AI landscape. Developers worldwide can now harness the power of GPT-4 and the suite of advanced APIs to create groundbreaking applications and engage users in dynamic and interactive conversations. As OpenAI continues to invest in chat completions and pushes the boundaries of AI, the future holds immense promise for developers seeking to leverage the capabilities of these cutting-edge language models.

https://openai.com/blog/gpt-4-api-general-availability

{
"prompt": "GPT-4 API Released, Marking the Deprecation of Older Models in Completions API, deepleaps.com, Photo, Cinematic, Surrealist",
"seed": 3555027762,
"used_random_seed": true,
"negative_prompt": "",
"num_outputs": 1,
"num_inference_steps": 25,
"guidance_scale": 7.5,
"width": 512,
"height": 512,
"vram_usage_level": "balanced",
"sampler_name": "euler",
"use_stable_diffusion_model": "Dreamshaper_3.32_baked_vae_clip_fix",
"clip_skip": false,
"tiling": "none",
"use_vae_model": "vae-ft-mse-840000-ema-pruned",
"stream_progress_updates": true,
"stream_image_progress": false,
"show_only_filtered_image": true,
"block_nsfw": false,
"output_format": "jpeg",
"output_quality": 75,
"output_lossless": false,
"metadata_output_format": "json",
"original_prompt": "GPT-4 API Released, Marking the Deprecation of Older Models in Completions API, deepleaps.com",
"active_tags": [
"Photo",
"Cinematic",
"Surrealist"
],
"inactive_tags": [],
"use_upscale": "RealESRGAN_x4plus",
"upscale_amount": "4"
}