Empero AI

2026•Empero AI

Qwythos-9B v2

Qwythos-9B-v2-GGUF is a quantized collection of the Qwythos-9B-v2 multimodal model, optimized for GGUF runtimes like llama.cpp and Ollama. This version specifically addresses looping behaviors found in the base model using Final-Token Preference Optimization (FTPO) while restoring the native multi-token prediction (MTP) head for improved performance.

Looping/degeneration eliminated with 0% repetition rate

Restored native multi-token prediction (MTP) head for speculative decoding

Multimodal vision capabilities via Qwen3.5 architecture

PricingFree

Version2

2026•Empero AI

Qwythos-9B v2

Open Source

gguf

llama.cpp

Qwythos-9B-Claude-Mythos-5-1M-GGUF is a 9-billion parameter reasoning model developed by Empero AI. It is post-trained on over 500 million tokens of high-quality Claude Mythos and Claude Fable traces, utilizing Empero AI's internal 'rethink' tool for chain-of-thought generation. This model outperforms the base Qwen3.5-9B model, achieving a +34 point improvement on the MMLU benchmark, +30 points on gsm8k-strict, and +19 points on gsm8k-flex. It supports native function calling as per the Qwen3.5 specification and offers a 1,048,576-token (1M) context window through YaRN rope-scaling. The model is available under the Apache-2.0 license and is compatible with various GGUF runtimes, including llama.cpp, Ollama, LM Studio, jan, and KoboldCpp.

9-billion parameter reasoning model

Post-trained on over 500 million tokens of high-quality Claude Mythos and Claude Fable traces

Outperforms base Qwen3.5-9B model on multiple benchmarks

PricingFree

Version2