SOURCE-REVIEWED MODEL PROFILE
Qwen3.8-Flash-Next
A multimodal sparse model combining linear and sparse attention, mixture-of-experts routing, n-gram embeddings and multi-token prediction.
Metadata reviewed 2026-10-24 · Original creator: Qwen
What it does
- 125B language-model parameters; 6B activated
- Additional 51B n-gram embeddings and 4B MTP
- Vision encoder and long-context architecture
Local fit
Large RAM / specialized offload engine. Our tests use a separate pruned Coder IQ1_M artifact through Strata at 16K context with vision disabled. Those measurements do not benchmark these original full weights.
Publisher/model configuration specification. Usable context depends on deployment resources.
Evidence and limits
Source-reviewed research profile; this exact checkpoint has not been benchmarked locally.
Research profile only
License and source
Custom community license, not MIT. Review its conditions and the separate quantization’s terms.
Read the original model card ↗
Reviewed source revision: de4b8e4d43b917e7706784d8bb445c9af86a3540