Q4_K_M · 4K context
Qwen3.8-Flash-Next MoE model with 125B total parameters and 6B active per token · estimated Q4_K_M memory about 108GB
Approximately 180B resident weight MoE model consisting of 125B core LMs (6B active) with 51B n-gram/PLE embeddings and 4B MTPs built into the architecture. The built-in 51B PLE embedding table is RAM offloadable and is independent of the Prompt Lookup speculative decoding that looks for repeated prompt phrases.
- Estimated Q4_K_M memory
- about 108GB
- total parameters
- 125B
- active parameter
- 6B
- native context
- 256K
Start with a model
Find hardware for this model
Also running document search?
Reserve this for embedding and reranking models sharing the same GPU or unified memory. Leave it at none if they run separately on the CPU.
Matching devices 29· 5 shown
Comfortable fit · Q4_K_M · Headroom 81.0GB
₩19,250,000
115.0 ~ 172.6 tok/s
Comfortable fit · Q4_K_M · Headroom 124.9GB
₩19,818,800
16.7 ~ 33.8 tok/s
Comfortable fit · Q4_K_M · Headroom 372.9GB
₩25,000,000
50 ~ 70.3 tok/s
Comfortable fit · Q4_K_M · Headroom 325.0GB
₩38,475,000
103.5 ~ 184.1 tok/s
Can run · check speed · IQ2_XS · Headroom 0.0GB
₩1,863,585
33.5 ~ 73.9 tok/s
Price is the range midpoint, with a basic host added for GPU-only prices. Speed is an estimate for the same model and 4K input.
Qwen3.8-Flash-Next Local Serving Run Recipe
Selecting a device changes the quantization, safety context, GPU offload, Flash Attention, KV cache, and batch start values.
If you want to choose equipment again
You can first narrow down the model configurations based on budget, noise level, operating system and whether it's acceptable used.