Qwythos: 9B Model with 1M Token Context from Empero

Empero has released Qwythos-9B, a model based on Qwen3.5-9B, trained on 500 million tokens of synthetic reasoning with support for a 1 million token context window.

Compiled by Sergey KostenchukPublished 2026-06-24Updated 2026-06-24

2026-06-24 Research Anthropic

🤖 Qwythos: 9B Model with 1M Token Context from Empero

The Qwythos-9B model has been released, based on the uncensored Qwen3.5-9B. It underwent full fine-tuning on 500 million tokens of synthetic reasoning (Chain of Thought) derived from Fable-5 and Mythos-5 logs. Thanks to YaRN technology and the Gated-DeltaNet architecture, the model supports a context window of 1 million tokens with sub-quadratic memory growth.

🌍 Demonstrates the effectiveness of using synthetic CoT data to boost the logic of small models to the level of large systems.

👤 A tool for working with massive volumes of data (code, documents) on local hardware.

Source 1: https://huggingface.co/empero-ai/Qwythos-9B-Claude-Mythos-5-1M

Sources