Run Qwen3.8 Flash Next on Apple Silicon
Qwen3.8 Flash Next is a mixture of experts model from Alibaba's Qwen team, released in August 2026: 125B parameters with 6B active per token, plus a 51B n gram embedding table and a 4B multi token prediction layer, about 180B in all, with a 256K context and image input. At Q4_K_M the weights are about 109 GB, too tight for a 128 GB Mac once the runtime and context are added, so a Mac Studio M3 Ultra with 256 GB is the comfortable single machine. Only 6B parameters are read for each token, so it generates fast for its size.
How much memory does Qwen3.8 Flash Next need?
Weight size is parameter count × bits per weight; recommended memory adds 15% runtime headroom plus a KV-cache allowance for the 262,144-token context window.
| Quantization | Weights | Recommended RAM | Cheapest fit | Notes |
|---|---|---|---|---|
| Q4_K_M | ~109.1 GB | 128 GB | M4 Max (Mac Studio) | Best quality/size trade-off — the default for most deployments |
| Q8_0 | ~191.3 GB | 224 GB | M3 Ultra (Mac Studio) | Near-lossless; noticeably better on code and long reasoning |
| FP16 | ~360 GB | 416 GB | Thunderbolt 5 cluster | Full precision — research and eval baselines |
Which Mac should you pick?
Generation speed on Apple Silicon scales mostly with memory bandwidth, so the chip choice sets both whether the model fits and how fast it runs.
| Chip | Max RAM | Bandwidth | Fits Qwen3.8 Flash Next? | From |
|---|---|---|---|---|
| M4 · Mac mini | 32 GB | 120 GB/s | Cluster only | $99/mo |
| M4 Pro · Mac mini | 64 GB | 273 GB/s | Cluster only | $200/mo |
| M4 Max · Mac Studio | 128 GB | 546 GB/s | Q4 | $286/mo |
| M3 Ultra · Mac Studio | 256 GB | 819 GB/s | Q4 + Q8 | $571/mo |
Deploy Qwen3.8 Flash Next in about 5 minutes
- 1
Build a Mac in the constructor — pick an M4 Max with at least 128 GB of unified memory.
- 2
Pick any Ollama-based stack (or a clean machine) and pull Qwen3.8 Flash Next with one command — every deployment ships the OpenAI-compatible API.
- 3
Point your existing OpenAI SDK code at the new base_url and ship.
from openai import OpenAI
client = OpenAI(
api_key="mcy_live_...",
base_url="https://dep-xxxx.macyou.cloud/v1",
)
response = client.chat.completions.create(
model="qwen-3-8-flash-next",
messages=[{"role": "user", "content": "Hello!"}],
)License and what it's good at
Qwen Community License 1.0 — Qwen Community License 1.0: commercial use is allowed, but hosted model services and products built mainly for AI coding or office work need a separate license from Qwen; internal use is exempt. Made by Alibaba (Qwen).
- Mixture of experts: 6B active parameters per token
- Hybrid linear attention keeps the KV cache small at long context
- Text and image input with a 256K context
Typical workloads: fast private assistants on a single large mac; long document work where kv cache memory matters; internal tools that stay inside the company.
Frequently asked questions
Can a 128 GB Mac run Qwen3.8 Flash Next?
Not comfortably at Q4_K_M: the weights alone are about 109 GB, and the runtime and macOS need room too. A Mac with 256 GB runs it with space for long context.
Can I use it in a commercial product?
Usually yes, but the Qwen Community License 1.0 asks hosted model services and products built mainly for AI coding or office productivity to get a separate license from Qwen. Internal use is exempt. Read the license before you ship.
Is it in Ollama?
Yes, Ollama lists qwen3.8 flash next builds, and community GGUF builds exist for llama.cpp. Qwen's own card recommends SGLang and vLLM for serving.