tawkeed-sa/tawkeed-gpt

TEXT GENERATIONConcurrency Cost:3Model Size:35.1BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 17, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Tawkeed GPT is a 35.1 billion parameter language model developed by Tawkeed-sa, forked from nex-agi/Nex-N2-mini and built upon the Qwen3.5-35B-A3B-Base architecture. This model maintains the Qwen3.5 MoE lineage, providing a robust foundation for various language tasks. It is designed for general-purpose applications, leveraging its base model's capabilities.

Loading preview...

Overview

Tawkeed GPT is a 35.1 billion parameter language model maintained by Tawkeed-sa. It is a direct fork of the nex-agi/Nex-N2-mini model, which itself is built upon the Qwen3.5-35B-A3B-Base architecture. This model explicitly retains the Qwen3.5 MoE (Mixture of Experts) lineage, ensuring a strong foundation for its capabilities. The model is licensed under Apache 2.0.

Key Characteristics

  • Architecture: Based on the Qwen3.5 MoE / qwen3_5_moe architecture.
  • Lineage: Directly derived from nex-agi/Nex-N2-mini, which uses Qwen3.5-35B-A3B-Base as its upstream base.
  • Parameters: 35.1 billion parameters.
  • Context Length: Supports a context length of 32768 tokens.

Usage and Future Development

Currently, Tawkeed GPT is a branded fork without additional Tawkeed-specific post-training. However, the repository is set up to accommodate future fine-tuning or continued post-training by Tawkeed, with provisions to update the model card with new training details and merged checkpoints or adapters. Developers can use the transformers library to interact with the model, as demonstrated in the provided Python example for chat template application and text generation.

Finetunes
1 models