Qwen 3.8 Flash Next (180B-A6B, 4-bit)
Qwen's experimental preview of the architecture behind Qwen4: a 125B-parameter MoE with 6B active per token, plus a 51B n-gram embedding table. Gezel offers the 94 GB Unsloth UD-IQ4_XS split GGUF for llama.cpp and the verified 112 GB native MLX Community group-32 conversion for Apple silicon. With mlx-vlm 0.7.1, Gezel memory-maps the MLX build's 29.80 GiB PLE table instead of wiring it into Metal; the measured fixed working set is about 75.2 GiB after the first live batch, and a 128 GB Mac is recommended.
| At a glance | |
|---|---|
| Made by | Qwen |
| Download size | 93.7 GB |
| Parameters | 180B |
| Quantization | UD-IQ4_XS |
| Context window | 262K tokens |
| Uses tools | Yes |
| Best for | Everyday work |
| Runs on | llama.cpp, MLX |
| License | Qwen Community 1.0 |
| Added to the catalog | 2026-09-17 |
Gezel downloads this model the first time a gezel needs it, checks the file against its published hash, and applies tuned settings for it.