Tokenai

13/06/2026

We are excited to announce the release of NeuralNode 2.2.0 on PyPI, bringing significant enhancements and major new capabilities to our flagship Python framework.

This release introduces deep integration with new AI providers, including Grok, OpenRouter, and Cohere, expanding the range of models and services available to developers.

NeuralNode 2.2.0 also delivers full integration with OpenClaw. Developers can now download, configure, customize, and manage OpenClaw directly through NeuralNode, providing complete control over deployment and ex*****on workflows from a single framework.

In addition, we have introduced a brand-new WhatsApp API integration, enabling users to communicate with their AI agents directly through WhatsApp and create more accessible conversational experiences.

Alongside this release, we are launching the official NeuralNode Platform, featuring both an AI Chatbot Builder and a Blueprint System. These tools allow anyone to build and deploy intelligent AI agents with zero programming experience. Users can simply describe their requirements to the chatbot or use the Blueprint system to create fully customized AI agents tailored to their needs.

Explore everything new in NeuralNode 2.2.0 and discover the official platform at:

https://neuralnode.tokenai.cloud/

11/06/2026

We're officially on X!

Follow us for the latest updates and announcements from TokenAI.

https://x.com/tokenaia

04/06/2026

Today, we are announcing the launch of Horus Lens, a new series of models within the Horus family dedicated to text-to-image generation.

Horus Lens expands the Horus ecosystem with specialized AI models designed specifically for image creation from text prompts.

As the first release in this new series, we are introducing
Horus Lens 1.0, a powerful image generation model available in five different versions.

We believe the Horus Lens series will be a significant addition to the AI ecosystem, contributing to the advancement of artificial intelligence innovation in Egypt and across the Middle East.

The release includes the full-precision model alongside four compressed GGUF variants, providing users with flexible deployment options across a wide range of hardware environments.

Horus Lens 1.0 is fully open source under the Apache License 2.0 and is available through our official AI framework, NeuralNode.

Official model page:
https://tokenai.cloud/models/horus-lens-1-0

We look forward to seeing what developers, researchers, and creators build with Horus Lens.

03/06/2026

Today, we are proud to announce a new research publication by the TokenAI Research Team titled "AI Comes at a Cost".

This paper examines more than 25 critical challenges and side effects associated with artificial intelligence, focusing on the impact of excessive AI usage on human cognition, behavior, and overall well-being. It provides a comprehensive analysis of how prolonged dependence on AI systems can influence individuals across multiple dimensions.

The research paper is now available on SSRN and can be accessed

through the following link:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6811419

30/05/2026

We are proud to announce the beginning of development for Horus Taleeq, a language model from the Horus family of models, specialized in more than ten different Arabic dialects, with a primary focus on the Arabic language and its regional dialects.
Horus Taleeq is being developed as a compact Small Language Model (SLM), optimized specifically for Arabic and its diverse dialects. Despite its relatively small size, the model is designed to outperform significantly larger models often many times its size across Arabic language benchmarks and dialect understanding tasks.
Horus Taleeq is being developed to become one of the world’s leading AI language models for Arabic and Arabic dialects. The model is planned to be officially released in the coming months as part of our long-term vision to expand the Horus family of models and contribute to the advancement of artificial intelligence in the Middle East.

12/05/2026

Today, our research paper
“Stable Training with Adaptive Momentum (STAM)"
was officially published on SSRN

You can review the paper through the following link:
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6699059

08/05/2026

Today, we announce a new research paper from the TokenAI research team Stable Training with Adaptive Momentum (STAM).
The paper introduces the STAM optimizer and its lightweight variant, STAMLite, focusing on improving training stability and performance under challenging conditions such as non-stationary data and long-horizon learning.
For more details, please visit:
https://tokenai.cloud/research/stam

23/04/2026

We are pleased to announce a strategic partnership with FutureTeach Academy to launch the Perception26 Program, marking its first edition in 2026.

The program aims to bring together developers and tech talents to collaborate on impactful, real-world AI projects under the umbrella of TokenAI, enabling participants to gain practical experience and build a strong professional foundation.

Through the Perception26 Program, participants will have the opportunity to actively contribute to real AI projects, collaborate in developing and refining cutting-edge models, and publish and deploy projects under their own names. This experience is designed to help participants build a professional portfolio with tangible, real-world outcomes.

The program offers a comprehensive experience, including:

* Mentorship from industry experts and researchers
* Flexible remote work environment
* Certificate of completion and recommendation letter
* Networking opportunities within the AI community

This program is designed for individuals passionate about advancing their skills and building a strong presence in artificial intelligence through hands-on experience.

Apply now: https://tokenai.cloud/join

20/04/2026

Today, we are releasing the source code used in the development and training of the Horus-1.0-4B model as open source.

This initiative is aimed at empowering developers, researchers, and anyone interested in advancing the model or leveraging its architecture for their own innovations.

You can access the full source code on GitHub

Contribute to tokenaii/horus-1.0 development by creating an account on GitHub.

08/04/2026

We’re happy to announce the launch of our Horus 1.0 open-source models family.
Today, we are officially releasing Horus-1.0-4B, the first public model in the Horus series — a multilingual text-generation model trained entirely from scratch on trillions of clean training tokens.
Horus-1.0-4B features an 8K context length and is available in 7 different versions, including the full-weights release and 6 compressed variants designed for flexible local deployment.
The model demonstrates strong reasoning, chain-of-thought, and thinking capabilities, achieving competitive benchmark results across MMLU and MMLU Pro, outperforming several widely used global models in its class.
Horus is fully open source and now available through TokenAI, with easy local deployment via neuralnode in Python.
This release marks an important step toward building real
open-source AI infrastructure from Egypt for the global AI community.
Explore the model and all available versions:
https://tokenai.cloud/horus