PixelVerse t1

Go to changelog

OpenAI has announced o1, a series of AI models that are designed to think more deeply and comprehensively before responding to prompts. This enhanced reasoning ability allows it to tackle complex problems, solve intricate puzzles, and even outperform human experts in certain domains such as coding, medical and mathematics.

Today, we are announcing our first ever AI models, designed to think logically using the Chain of Thought prompting strategy similar to o1. Our research indicates that by prompting AI models to not only reason logically but also express and demonstrate their thought processes, their intelligence and problem-solving abilities are significantly enhanced.

Meet the t1 Series

Today, we are excited to announce the series of PixelVerse t1 models:

• t1 mini - powered by llama 3.1 70b & 3.2 90b, 3b, 2b and more coming soon
• t1 core - powered by llama 3.1 401b (COMING SOON)
• t1 pro - powered by gpt 4o (COMING SOON)

While the base models provide a strong foundation, our custom prompts are designed to enhance the AI's ability to understand and express and demonstrate its logical thinking and reasoning. In our tests, this approach significantly elevates the intelligence and quality of responses compared to the base models by more than 50% as shown below in our t1-mini comparisons.

t1-mini Comparisons

t1 scores exceptionally well compared to other models, such as GPT 4o, Llama 3.1 & Gemini 1.5 Pro and Anthropic 3.5 Sonnet.

Here’s some benchmarks for the t1-mini model with the latest and most advanced AI models in the world, GPT 4o, Llama 3.1, Gemini 1.5 Pro and Anthropic 3.5 Sonnet:

Alternately, you can view more comparisons on our X post:

Don't have access to @OpenAI's o1?

Meet PixelVerse t1 Mini, a powerful logical reasoning model based on Llama that solves advanced maths, science and language problems.

Try it today for free: https://t.co/5FLyhn7EaG

🧵Thread on comparison with ChatGPT, Meta & Gemini: pic.twitter.com/jHM7jigUWR

— PixelVerseIT (@pixelverseIT) September 14, 2024

Note that these questions don’t test all of the AI model’s capabilities. It also doesn’t necessarily mean that t1 mini is better than those models listed above in our comparisons, but instead demostrates that prompt techniques alone can allow an open source model to significantly increase in quality.

We will be publishing t1-core and t1-pro model benchmarks and comparisons soon. You can contribute to the development of t1 and future models by providing difficult questions or prompts by filling out this form. The more difficult the questions, the better!

t1 Access

You can try out t1-mini today on any device with a browser on PixelVerseAI Labs for free. We will be pushing out new updates and features in the coming weeks and listening to your feedback. If you have any questions or need assistance, please contact our team.

t1 API

This is just a preview of our logical thinking and reasoning models in AI Labs and PixelVerse Chats. We will be adding new features soon, such as file and image uploading and prompt updates to increase the capabilities of t1.

In addition, we will be releasing our t1 API to the general public in the coming weeks. You can learn more and join the waitlist below:

We are excited on what’s next for AI and hope to bring even better and useful models to you soon!

t1 Changelog

Here are the latest updates and changes to the t1 models. Please note that not all changes are listed here, but only the major ones.

• v0.4.0 - 07 October 2024 - Added thumbs up/down feedback buttons for each t1 response along with copy buttons. Voice speaking icon added and functionality coming soon.
• v0.3.0 - 06 October 2024 - Updates to t1 model for better responses.
• v0.2.0 - 29 September 2024 - Added new Meta Llama 3.2 models (90b, 3b & 2b) and Gemma 2 7b to t1. Also made UI improvements.
• v0.1.5 - 17 September 2024 - Further Enhanced UI
• v0.1.0 - 17 September 2024 - Improved UI and removed junky UI
• v0.0.3 - 16 September 2024 - Improved t1 model responses with CoT
• v0.0.1 - 14 September 2024 - t1 mini released on PixelVerse AI Labs