* This blog post is a summary of this video.

OpenAI's Groundbreaking GPT4 Turbo and GPT Store Announcements

Author: Matt WolfeTime: 2024-01-29 09:50:00

Table of Contents

GPT4 Turbo Brings Major Upgrades for Developers

OpenAI hosted their first ever developer conference and announced GPT4 Turbo, bringing major upgrades for developers building with their API. GPT4 Turbo includes a 128,000 token context window for more knowledge retention, JSON mode for reproducible outputs, new modalities like DALL-E images and Whisper speech recognition, lower costs down to 1 cent per 1000 tokens, and more.

These upgrades provide developers more control, functionality, and affordability when building AI applications powered by OpenAI's API. The expanded context window allows models to remember more information and have more knowledgeable conversations. JSON mode ensures predictable structured responses from the API. And the addition of new modalities opens up more possibilities for multi-modal AI.

Larger Context Windows

One of the biggest upgrades in GPT4 Turbo is expanding context length from 8,000 tokens up to 128,000 tokens. This allows the model to remember up to 96,000 words of conversation history and continue discussions with more background knowledge. For comparison, this is over 20,000 more tokens than Anthropic's Claude model. So GPT4 Turbo has the largest publicly available context window of any model, enabling more detailed and contextual conversations.

More Control and Reproducible Outputs

GPT4 Turbo also provides more control with features like JSON mode. This ensures the model will respond with valid JSON structured data, making API integrations easier. Developers also have more control over model behavior with the new reproducible outputs feature. By passing a seed parameter, the model will return consistent outputs given the same inputs. Additionally, GPT4 Turbo has better function calling capabilities. Developers can invoke multiple functions at once, and the model follows instructions more accurately.

New Modalities

GPT4 Turbo also introduces new modalities beyond just text. Developers now have access to DALL-E for AI image generation directly through the API. They can provide an image prompt and generate captions, classifications and analysis. There is also a new state-of-the-art text-to-speech model with natural sounding voices. Developers can input text and get back high quality synthesized speech.

Lower Costs

Finally, GPT4 Turbo comes with lower costs. Prompt tokens are now just 1 cent per 1,000, while completion tokens are 3 cents per 1,000. This means costs are reduced by 3X for prompt tokens and 2X for completion tokens compared to the previous GPT-4 pricing. The lower costs allow developers to build and scale AI applications more affordably than ever before. And OpenAI is also doubling token allotments for existing GPT-4 customers.

GPT Store Opens Up Sharing and Selling Custom AI

Beyond the GPT4 Turbo API updates, OpenAI also announced the GPT Store for sharing and even selling customized AI models known as GPTs. The GPT Store provides an easy way for anyone to build a custom GPT with instructions, knowledge and capabilities. Popular GPTs can then be published to share or sell to others.

For example, developers could create a startup advisor GPT by uploading transcripts and lectures as knowledge. The published GPT would then leverage this domain expertise to provide specific advice. Creators of popular GPTs in the store even earn a revenue share from OpenAI.

Build GPTs in Natural Language

The key innovation with GPTs is the natural language builder interface inside ChatGPT. Users can simply describe what they want the GPT bot to do, and ChatGPT will guide them through setting up instructions, knowledge and capabilities. For example, saying "I want to build a bot to optimize Twitter posts" will kick off a conversational process to create that custom AI assistant.

Customize and Monetize

By customizing capabilities, data and behaviors, GPT builders can create uniquely valuable bots. The startup advisor example shows the power of domain-specific knowledge injection. Other GPTs may focus on specific actions or integrations. Popular published GPTs can even earn their builders a cut of usage revenue from OpenAI. This provides an incentive for creators to build genuinely useful AI tools.

Experiment with GPT4 Turbo in Playground

While the full GPT4 Turbo and GPT Store aren't available to all yet, developers can start experimenting with the key features in OpenAI's developer Playground today. The Playground provides access to models like GPT4 Turbo for testing interactions.

Key capabilities like persistent memory and knowledge injection are available in Playground's Assistant feature. This allows developers to preview how they may leverage things like context and documents in their own applications.

Persistent Memory

One option available with Assistants is Threads, which provide persistent memory across conversations. The context allows more natural back-and-forth dialog compared to isolated stateless queries. Developers can see how this context shifts responses as more of a discussion occurs. Testingassisted conversations in Playground provides insight into the possibilities with rememberable bots.

Inject Knowledge

Playground Assistants also showcase the power of injected knowledge through file uploads. Developers can feed documents like PDFs to customize model understanding. Seeing examples of ingested documents affecting responses demonstrates how developers may leverage external knowledge sources in future applications.

The Future of AI Development

With tools like GPT4 Turbo, the GPT Store and Playground, OpenAI is providing developers more access, control and customization over conversational AI than ever before. The possibilities span from multifunctional virtual assistants to domain-specific advisors.

However, these rapid innovations also disrupt existing businesses built on top of OpenAI APIs. Any successes get copied into core platform offerings. So developers should focus on unique value rather than easily replicable demos.


Q: What is GPT4 Turbo?
A: GPT4 Turbo is the latest upgraded version of OpenAI's generative AI model, bringing major improvements like larger context windows up to 128,000 tokens.

Q: How can I access GPT4 Turbo features?
A: You can experiment with GPT4 Turbo now using OpenAI's Playground if you have an API key. GPT4 Turbo upgrades are also rolling out to ChatGPT.

Q: What is the GPT Store?
A: The GPT Store is OpenAI's new marketplace for sharing and selling custom-built GPTs using natural language instructions.

Q: Can I make money selling GPTs?
A: Yes, OpenAI will pay creators of the most popular and useful GPTs listed in the store through a revenue sharing program.

Q: How do custom assistants in Playground work?
A: You can create AI assistants with persistent memory to have contextual conversations using prompts and uploaded documents.