Page History

• Latest model
• With broad general knowledge and domain expertise, GPT-4 can follow complex instructions in natural language and solve difficult problems with with greater accuracy.

• is more creative and collaborative than ever before. It can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user’s writing style.

• Following the research path from GPT, GPT-2, and GPT-3, the deep learning approach leverages more data and more computation to create increasingly sophisticated and capable language models

• 6 months were spent making GPT-4 safer and more aligned.

• GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations.

• Price list for GPT-4  (Multiple models, each with different capabilities and price points. Prices are per 1,000 tokens. You can think of tokens as pieces of words, where 1,000 tokens is about 750 words.)

  Model	Prompt	Completion
  8K context	$0.03 / 1K tokens	$0.06 / 1K tokens
  32K context	$0.06 / 1K tokens	$0.12 / 1K tokens

• GPT-4 models

  LATEST MODEL	DESCRIPTION	MAX TOKENS	TRAINING DATA
  gpt-4	More capable than any GPT-3.5 model, able to do more complex tasks, and optimized for chat. Will be updated with our latest model iteration.	8,192 tokens	Up to Sep 2021
  gpt-4-0314	Snapshot of gpt-4 from March 14th 2023. Unlike gpt-4, this model will not receive updates, and will only be supported for a three month period ending on June 14th 2023.	8,192 tokens	Up to Sep 2021
  gpt-4-32k	Same capabilities as the base gpt-4 mode but with 4x the context length. Will be updated with our latest model iteration.	32,768 tokens	Up to Sep 2021
  gpt-4-32k-0314	Snapshot of gpt-4-32 from March 14th 2023. Unlike gpt-4-32k, this model will not receive updates, and will only be supported for a three month period ending on June 14th 2023.	32,768 tokens	Up to Sep 2021

  
  

  For many basic tasks, the difference between GPT-4 and GPT-3.5 models is not significant. However, in more complex reasoning situations, GPT-4 is much more capable than any of our previous models.

• Limitation:

          GPT-4 is currently in a limited beta and only accessible to those who have been granted access. In order to use this API we need to join the waitlist to get access when capacity is available.

GPT-3.5 models can understand and generate natural language or code. Our most capable and cost effective model in the GPT-3.5 family is gpt-3.5-turbo which has been optimized for chat but works well for traditional completions tasks as well.

Models referred to as "GPT 3.5"

GPT-3.5 series is a series of models that was trained on a blend of text and code from before Q4 2021. The following models are in the GPT-3.5 series:

1. 1. code-davinci-002 is a base model, so good for pure code-completion tasks
   2. text-davinci-002 is an InstructGPT model based on code-davinci-002
   3. text-davinci-003 is an improvement on text-davinci-002
   4. gpt-3.5-turbo-0301 is an improvement on text-davinci-003, optimized for chat

Recommendation to use gpt-3.5-turbo over the other GPT-3.5 models because of its lower cost.

Experimenting with gpt-3.5-turbo is a great way to find out what the API is capable of doing. After you have an idea of what you want to accomplish, you can stay with gpt-3.5-turbo or another model and try to optimize around its capabilities.

Note: OpenAI models are non-deterministic, meaning that identical inputs can yield different outputs. Setting temperature to 0 will make the outputs mostly deterministic, but a small amount of variability may remain.

• GPT-3 models can understand and generate natural language.
• These models were superceded by the more powerful GPT-3.5 generation models.
• However, the original GPT-3 base models (davinci, curie, ada, and babbage) are current the only models that are available to fine-tune.
• Fine-tuning mean to build and train our own data from one of GPT-3 Models
• Prices:

• GPT-3 Doc

• Enhanced Capabilities of davinci-text-003

The main difference between davinci-text-002 and davinci-text-003 is that the latter has been trained on a larger dataset, allowing it to generate more accurate results. Additionally, text-davinci-003 can better understand natural language instructions and produce more detailed responses. It also has improved capabilities for summarizing long documents and generating coherent paragraphs from multiple sources. Furthermore, OpenAI claims that text-davinci-003 can generate longer content with greater clarity, and engagement.

• Strengths and Weaknesses of Each Model

The output from davinci-text-002 was a series of short, concise sentences that were easy to read and understand. On the other hand, davinci-text-003 produced longer, more complex sentences with more intricate language. Both outputs had their own unique style and could be used for different purposes depending on the context. It is clear that both algorithms have their strengths and weaknesses when it comes to text generation. Note: ‘003 takes longer than the previous version to process prompts.

• text-davinci-003 includes the following improvements:

  • It produces higher quality writing. This will help your applications deliver clearer, more engaging, and more compelling content.
  • It can handle more complex instructions, meaning you can get even more creative with how you make use of its capabilities now.
  • It’s better at longer form content generation, allowing you to take on tasks that would have previously been too difficult to achieve.

  Expand
  title To illustrate the improvements, check out the following example:
  text-davinci-002: Prompt: "I am trying to write a web application. What backend frameworks could I use, and what are their pros and cons?" There are many backend frameworks that could be used for a web application. Some popular one text-davinci-003: Prompt: "I am trying to write a web application. What backend frameworks could I use, and what are their pros and cons?" 1.  Node.js: Node.js is a JavaScript-based runtime environment that is popular for writing backend applications. Pros: Node.js is fast, lightweight, and easy to learn, and it has a large community of developers. Cons: Node.js can be difficult to debug and there is a lack of robust tools for building large-scale applications. 2.  Django: Django is a popular Python-based web framework. Pros: Django is powerful and well-documented, making it easy to get started. It also supports a wide range of databases and APIs, allowing you to easily integrate with other services. Cons: Django can be difficult to scale for large applications and can be slow for certain types of requests.3.  Ruby on Rails: Ruby on Rails is a popular web framework based on the Ruby programming language. Pros: Ruby on Rails is easy to learn and can be used to quickly build powerful web applications. It also has a wide range of built-in features, such as authentication and authorization. Cons: Ruby on Rails can be slow and resource-intensive, and it can be difficult to scale. text-davinci-003 can be accessed through openai API and Playground at the same price as other Davinci base language models.

  • Other examples of outputs for each model for the same prompt

  Users can access `text-davinci-003` through the Open AI API and playground, at the same price as their other Davinci base language models ($0.0200 / 1k tokens). 

• OpenAI's new GPT-3.5 Turbo model offers a cost-effective option for many non-chat use cases.
• While GPT-3.5 Turbo performs well in 0-shot classification and math, Davinci-003 performs slightly better in k-shot classification and may be a better option for those looking for clear, concise responses that get straight to the point.
• Comparison for different tasks, can be found here

  • Expand
    title Classification
    From our previous experiments, we concluded that Davinci-003 performed much better than Davinci-002 with 0 shot prompts (92% vs 69.91%), and on par or slightly worse than Davinci-002 on few shot prompts (87.2% vs 91.6%). We observe a similar case with GPT-3.5 Turbo and Davinci-003. GPT-3.5 Turbo performs better on 0 shot classification- on an insider trading classification example it achieves 91.72% accuracy, versus 82.02% for Davinci-003.

  • Expand
    title Text Generation
    If you're looking for an AI language model that can generate long, detailed answers to complex questions, GPT-3.5 Turbo might seem like the obvious choice since it's been trained on a massive dataset of human language and produces coherent, contextually-appropriate responses. But what if you're looking for a model that can provide clear, concise responses? In that case, you might want to consider using Davinci-003 instead. In our experiment, we provided both GPT-3.5 Turbo and Davinci-003 with a set of 30 questions, and asked each model to provide 30 answers. What we found was that GPT-3.5 Turbo’s responses tended to be much longer than Davinci-003's, with many of them exceeding 100 words. In contrast, Davinci-003's responses were much more concise, often consisting of just a few sentences. Given a max tokens of 500, the average response from GPT-3.5 Turbo was 156 words (~208 tokens) nearly twice the amount of Davinci-003, whose responses were an average of 83 words (~111 tokens).

  • Expand
    title Math
    Notably, GPT-3.5 Turbo is significantly better than Davinci-003 at math. When evaluating by exact match with an answer key, GPT-3.5 Turbo gets the correct numerical answer 75% of the time, while Davinci-003 only gets 61% correct.

  • Overall, GPT-3.5 Turbo performs better than Davinci-003 on most tasks at 10% of the cost - but is much more rambly. You can test out whether GPT-3.5 Turbo is better for your use case through Spellbook - sign up today.

• GPT4 and GPT-3.5-turb API cost comparison and understanding
• Below you can find a list of references links:
  • https://me.pcmag.com/en/news/15528/gpt-4-vs-chatgpt-35-whats-the-difference
  • https://www.makeuseof.com/gpt-4-vs-gpt-35-differences-explained/
  • https://www.howtogeek.com/882274/gpt-3-5-vs-gpt-4/#:~:text=Compared%20to%20GPT-3.5%2C%20GPT-4%20is%20smarter%2C%20can%20handle,come%20with%20the%20hourly%20prompt%20restrictions%20GPT-4%20does.

• Expand
  title What Is GPT 3.5?
  GPT 3.5 is, as the name suggests, a sort of bridge between GPT-3 and GPT-4. OpenAI hasn’t really been particularly open about what makes GPT 3.5 specifically better than GPT 3, but it seems that the main goals were to increase the speed of the model and perhaps most importantly to reduce the cost of running it. Interestingly, what OpenAI has made available to users isn’t the raw core GPT 3.5, but rather several specialized offshoots. For example, GPT 3.5 Turbo is a version that’s been fine-tuned specifically for chat purposes, although it can generally still do all the other things GPT 3.5 can.

• Expand
  title What Is GPT 4?
  OpenAI’s GPT-4 has emerged as their most advanced language model yet, offering safer and more effective responses. This cutting-edge, multimodal system accepts both text and image inputs and generates text outputs, showcasing human-level performance on an array of professional and academic benchmarks. When comparing GPT-3 and GPT-4, the difference in their capabilities is striking. GPT-4 has enhanced reliability, creativity, and collaboration, as well as a greater ability to process more nuanced instructions. This marks a significant improvement over the already impressive GPT-3, which often made logic and other reasoning errors with more complex prompts.

  Expand
  title Other differences:
  Another key distinction between the two models lies in their size. GPT-3 boasts a remarkable 175 billion parameters, while GPT-4 takes it a step further with a (rumored) 1 trillion parameters. GPT-4is 10 times more advanced than its predecessor, GPT-3.5. This enhancement enables the model to better understand context and distinguish nuances, resulting in more accurate and coherent responses. Furthermore, GPT-4has a maximum token limit of 32,000 (equivalent to 25,000 words), which is a significant increase from GPT-3.5’s 4,000 tokens (equivalent to 3,125 words). While GPT-4 is not perfect, the measures it adopts to ensure safer responses are a welcomed upgrade from that of the GPT-3.5 model. With GPT-3.5, OpenAI took a more moderation-based approach to safety. In other words, some of the safety measures were more of an afterthought. OpenAI monitored what users did and the questions they asked, identified flaws, and tried to fix them on the go. One of GPT-3.5's flaws is its tendency to produce nonsensical and untruthful information confidently. In AI lingo, this is called "AI hallucination" and can cause distrust of AI-generated information. In GPT-4, hallucination is still a problem. However, according to the GPT-4 technical report, the new model is 19% to 29% less likely to hallucinate when compared to the GPT-3.5 model. But this isn't just about the technical report. Responses from the GPT-4 model on ChatGPT are noticeably more factual. A less talked about difference between GPT-4 and GPT-3.5 is the context window and context size. A context window is how much data a model can retain in its "memory" during a chat session and for how long. GPT-4 has a significantly better context size and window than its predecessor model. Another issue is the limitation on the volume of text you can use in a prompt at once. Summarizing long text using GPT-3 typically means splitting the text into multiple chunks and summarizing them bit by bit. The improvement in context length in the GPT-4 model means you can paste entire PDFs at a go and get the model to summarize without splitting it into chunks. GPT-4 has a much larger model size, which means it can handle more complex tasks and generate more accurate responses. This is thanks to its more extensive training dataset, which gives it a broader knowledge base and improved contextual understanding. GPT-4 is better equipped to handle longer text passages, maintain coherence, and generate contextually relevant responses. For this reason, it’s an incredibly powerful tool for natural language understanding applications. It’s so complex, some researchers from Microsoft think it’s shows “Sparks of Artificial General Intelligence” or AGI. But there’s a downside, as with any cutting-edge technology. The significant advancements in GPT-4 come at the cost of increased computational power requirements. This makes it less accessible to smaller organizations or individual developers who may not have the resources to invest in such a high-powered machine.

• Expand
  title When Should You Use GPT 3.5 or GPT 4?
  Whether you’re building your own GPT-based app or using ChatGPT and have a choice between GPT 3.5 and 4, there are some use cases where GPT 3.5, despite being objectively inferior to GPT-4, is a better choice: If you’re limited in terms of computing power, GPT 3.5 is significantly cheaper to run. Although GPT-4 is less likely to “hallucinate” or make factual errors than GPT 3.5, it takes the same amount of time for a human to fact-check and edit the text. So if speed is the priority, the older model might be the better choice. But there are good reasons to GPT 4 as well: If you need GPT to do anything that requires complex reasoning, you’ll want to use GPT-4 every time. If you need GPT to process more than 3000 words at once, you’ll have to go with GPT-4. Over time, as computing power becomes more powerful and less expensive, while GPT-4 and it’s successors become more efficient and refined, it’s likely that GPT-4 will replace GPT 3.5 in every situation. Until then, you’ll have to choose the model that best suits your resources and needs.

• Expand
  title More
  GPT-4 models (preview) GPT-4 can solve difficult problems with greater accuracy than any of OpenAI's previous models. Like gpt-35-turbo, GPT-4 is optimized for chat but works well for traditional completions tasks. These models are currently in preview. For access, existing Azure OpenAI customers can apply by filling out this form. gpt-4 gpt-4-32k The gpt-4 supports 8192 max input tokens and the gpt-4-32k supports up to 32,768 tokens. GPT-3 models The GPT-3 models can understand and generate natural language. The service offers four model capabilities, each with different levels of power and speed suitable 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: text-davinci-003 text-curie-001 text-babbage-001 text-ada-001 While Davinci is the most capable, the other models provide significant speed advantages. Our recommendation is for users to start with Davinci while experimenting, because it produces the best results and validate the value that Azure OpenAI can provide. Once you have a prototype working, you can then optimize your model choice with the best latency/performance balance for your application. So Davinci is the most capable model and can perform any task the other models can perform, often with less instruction. For applications requiring deep understanding of the content, like summarization for a specific audience and creative content generation, Davinci produces the best results. The increased capabilities provided by Davinci require more compute resources, so Davinci costs more and isn't as fast as other models. Another area where Davinci excels is in understanding the intent of text. Davinci is excellent at solving many kinds of logic problems and explaining the motives of characters. Davinci has been able to solve some of the most challenging AI problems involving cause and effect. Use for: Complex intent, cause and effect, summarization for audience

• Understand the pricing of GPT3

• ChatGPT API Pricing Comparison

• 

• Expand
  title The API is available on Microsoft Azure
  Azure customers can access the OpenAI API on Azure with the compliance, regional support, and enterprise-grade security that Azure offers. Learn more or contact sales@openai.com. What is Azure OpenAI Service? Here a link that shows Azure OpenAI Service pricing

The number of API tokens you will need to use the GPT-3 API in your app will depend on how frequently you plan to make API requests and how large those requests are.

OpenAI currently offers several different pricing tiers for API access, which are based on the number of tokens you purchase. The pricing tiers range from 400,000 tokens per month to 3 billion tokens per month, with the cost per token decreasing as you purchase larger quantities.

To estimate how many tokens you will need, you can start by calculating the average number of API requests your app will make per day, and the average size of those requests in terms of the number of tokens consumed per request. You can then use this information to determine how many tokens you will need on a monthly basis.

It's important to note that OpenAI has strict usage limits and guidelines for the GPT-3 API, and they reserve the right to suspend or terminate API access for users who violate these guidelines or exceed their usage limits. You should review the OpenAI API documentation carefully and ensure that your app complies with all usage guidelines and policies.

OpenAI offers a few different pricing plans for their API, and the cost per month will depend on the plan you choose and the number of tokens you purchase.

• The current pricing plans are as follows:
  • Developer plan: $60/month for 400,000 tokens
  • Team plan: $300/month for 3,000,000 tokens
  • Business plan: custom pricing for larger token amounts
• The cost per token decreases as you purchase larger quantities, and there are additional discounts available for annual commitments.
• You can purchase additional tokens through the OpenAI dashboard using a credit card or bank transfer.
• The exact payment methods available may vary depending on your location and the size of your purchase.
• It's important to note that these prices are subject to change, and you should consult the OpenAI website or contact their sales team for the most up-to-date pricing information.

• The pricing for the OpenAI API also depends on the model you use and the number of API requests you make.

• Each API request consumes a certain number of tokens, which are deducted from your account balance.

  The GPT-3 model is one of the most expensive models offered by OpenAI, with each API request consuming a minimum of 1 token per character generated. The exact number of tokens consumed per request will depend on the length and complexity of the generated text.

  The pricing plans for the OpenAI API are based on the number of tokens you purchase per month, and the cost per token decreases as you purchase larger quantities. The pricing plans also include a limit on the number of API requests you can make per month, which varies depending on the plan you choose.

  It's important to carefully consider your usage needs and choose a plan that provides enough tokens for your expected usage, as well as to monitor your usage to ensure you do not exceed your plan limits or usage guidelines.

Log in to your account to view your usage tracking dashboard. This page will show you how many tokens you’ve used during the current and past billing cycles.

You can configure a usage hard limit in your billing settings, after which we’ll stop serving your requests. You may also configure a soft limit to receive an email alert once you pass a certain usage threshold. There may be a delay in enforcing the limit, and you are responsible for any overage incurred. We recommend checking your usage tracking dashboard regularly to monitor your spend.

• Completions requests are billed based on the number of tokens sent in your prompt plus the number of tokens in the completion(s) returned by the API.
• The best_of and n parameters may also impact costs. Because these parameters generate multiple completions per prompt, they act as multipliers on the number of tokens returned.
• Your request may use up to num_tokens(prompt) + max_tokens * max(n, best_of) tokens, which will be billed at the per-engine rates outlined at the top of this page.
• In the simplest case, if your prompt contains 10 tokens and you request a single 90 token completion from the davinci engine, your request will use 100 tokens and will cost $0.002.
• You can limit costs by reducing prompt length or maximum response length, limiting usage of best_of/n, adding appropriate stop sequences, or using engines with lower per-token costs.