OpenAI has been updated heavily, and it has become stronger again! The API has a terrifying function call capability, can handle longer contexts, and the price has dropped by 75%!

On June 13, OpenAI's official website suddenly released the latest capability update of ChatGPT. The key points are as follows:

1. New function: A new function calling capability has been added to the Chat Completions API, which allows the model to call functions and generate corresponding JSON objects as output when needed. This enables developers to more accurately obtain structured data from models, realize the conversion from natural language to API calls or database queries, and can also be used to extract structured data from text.

(That is, after OpenAI’s fine-tuning, if you speak human words to it, it can recognize it and convert it into a function for you, further realizing the ability of programming without programming, and it is more convenient to obtain from the chaotic structure structured data)

2. Model update: updated more controllable gpt-4 and gpt-3.5-turbo versions, and a new 16k context version of gpt-3.5-turbo, which can handle longer texts than the standard 4k version.

(can support 20 pages of text!)

3. Model deprecation: Announced the deprecation timeline of the gpt-3.5-turbo-0301 and gpt-4-0314 models. Users of these models can choose to upgrade to the newer model until a certain time, after which the old model will no longer be available.

(On the one hand, pay attention to those who are still using these models, please keep up with the crazy OpenAI rhythm. On the other hand, many people who have benchmarked OpenAI’s previous version models, OpenAI has already abandoned them..)

4. Price adjustments: The price of the most advanced embedding model has been reduced by 75%, and the input token price of gpt-3.5-turbo has been reduced by 25%.

(Sam Altman said in the parade recently that the price will continue to drop, and this is coming. And the price drop is the ankle-cut price of the strongest model. The latest price, per 1k token, is 0.0001 US dollars)

OpenAI also emphasized that all of these models continue to maintain the data privacy and security guarantees introduced on March 1-clients own all the outputs they request to generate, and their API data will not be used for training. With these updates, we're inviting even more people on the waitlist to try GPT-4, and we look forward to seeing what projects you build with GPT-4! We encourage developer feedback to help us ensure a smooth transition of model updates.

The following is the full text of the announcement:

We released the gpt-3.5-turbo and gpt-4 models earlier this year, and in just a few months we've seen developers build incredible applications on top of these models. Today, we're following up with some exciting updates:

New function call functionality in the Chat Completions API

· Updated gpt-4 and gpt-3.5-turbo versions with added controllability

· New 16k context version of gpt-3.5-turbo (vs. standard 4k version)

75% reduction in the price of our state-of-the-art embedding model

25% reduction in the price of gpt-3.5-turbo input tokens

Announced deprecation timeline for gpt-3.5-turbo-0301 and gpt-4-0314 models

All of these models come with the same data privacy and security guarantees we launched on March 1st - customers own all output generated by their requests, and their API data is not used for training. function call

Developers can now describe functions to gpt-4-0613 and gpt-3.5-turbo-0613, and have the model intelligently choose to output a JSON object containing the arguments for calling those functions. This is a new way to more reliably connect GPT's capabilities with external tools and APIs. These models have been fine-tuned to both detect when a function needs to be called (depending on user input) and respond with JSON that conforms to the function signature. Function calls allow developers to more reliably retrieve structured data from models. For example, developers can:

Create chatbots that answer questions by calling external tools such as the ChatGPT plugin;

Turning a question like "Ask Anya if she wants coffee next Friday" into a function call like send_email(to: string, body: string), or "What's the weather in Boston right now?" into get_current _weather(location: string, unit: 'celsius' | 'fahrenheit');

· Convert natural language into API calls or database queries;

Convert "Who are my top ten customers this month?" into an internal API call such as get_customers_by_revenue(start_date: string, end_date: string, limit: int), or "on How many orders did Acme, Inc. place in a month?" Use sql_query(query: string) to convert to SQL query;

Extract structured data from text;

· Define a function called extract_people_data(people: [{name: string, birthday: string, location: string}]) to extract all mentions of people from a Wikipedia article.

These use cases are all enabled by new API parameters in our /v1/chat/completions endpoint, functions and function_call, which allow developers to describe functions to the model via JSON Schema, and optionally ask it to call specific functions. Please get started with our developer docs, and if you see situations where function calls could be improved, please add an evaluation.

function call example

What's the weather like in Boston right now?

step 1

OpenAI API

Invoke the model with the function and the user's input

step 2

Third-party APIs

Call your API with the model's response

step 3

OpenAI API

Send the response back to the model for summarization The weather in Boston right now is sunny with a temperature of 22 degrees Celsius. Since the alpha release of the ChatGPT plugin, we've learned a lot about how to make tools and language models work together safely. However, there are still some open research questions. For example, a proof-of-concept vulnerability illustrates how obtaining untrusted data from a tool's output can cause the model to perform unexpected actions. We are working hard to mitigate these and other risks. Developers can protect their apps by only using information from trusted tools and including user confirmation steps before taking actions that have real-world impact, such as sending an email, posting online, or making a purchase.

new model

GPT-4

gpt-4-0613 includes updated and improved models, and function call functionality.

gpt-4-32k-0613 includes the same improvements as gpt-4-0613, plus increased context length for better understanding of larger texts.

With these updates, we will be inviting more people on the waitlist to try out GPT-4 in the coming weeks, with the goal of removing the waitlist for this model entirely. Thanks to everyone who has been patient, we look forward to seeing what you build with GPT-4!

GPT-3.5 Turbo

gpt-3.5-turbo-0613 includes the same function call capabilities as GPT-4, as well as more reliable control through system messages, both of which allow developers to more efficiently guide model responses.

gpt-3.5-turbo-16k provides four times the context length of gpt-3.5-turbo, but at double the price: $0.003 per 1K input tokens and $0.004 per 1K output tokens. 16k context means models can now be used in About 20 pages of text are supported in one request.

model deprecation

Today we will begin the process of upgrading and deprecating for the initial versions of gpt-4 and gpt-3.5-turbo that we announced in March. Apps using the stable model names (gpt-3.5-turbo, gpt-4, and gpt-4-32k) will automatically upgrade to the new models listed above on June 27th. To compare performance between model versions, our s library supports public and private evaluations to demonstrate how model changes will affect your use case.

Developers who need more time to transition can continue to use the old model by specifying gpt-3.5-turbo-0301, gpt-4-0314, or gpt-4-32k-0314 in the 'model' parameter of their API requests . These old models will still be accessible after September 13th, after which requests specifying these model names will fail. You can stay up-to-date on model deprecations on our model deprecations page. This is the first update to these models; as such, we welcome developer feedback to help us ensure a smooth transition.

Lower the price

Effective today, we continue to improve the efficiency of our systems and pass on these cost savings to developers.

Embeddings

text-embedding-ada-002 is our most popular embedding model. Today we reduced its cost by 75% to $0.0001 per 1K tokens.

GPT-3.5 Turbo

gpt-3.5-turbo is our most popular chat model, providing ChatGPT service for millions of users. Today we reduced the input token cost of gpt-3.5-turbo by 25%. Developers can now use this model for $0.0015 per 1K input tokens and $0.002 per 1K output tokens, which equates to roughly 700 pages per dollar.

The price of gpt-3.5-turbo-16k will be $0.003 per 1K input tokens and $0.004 per 1K output tokens.

Developer feedback is the cornerstone of our platform evolution, and we will continue to make improvements based on the suggestions we receive. We look forward to seeing how developers use these latest models and new features in their apps.