Can I run Qwen 3.8 2.4T-A95B on a M6 16GB Mac?
Computed with the open FitLLM engine — accurate per-layer KV-cache modeling, not a naive estimate. Updated 2026-08-24.
Memory breakdown (8-bit, F16 KV, 33K context)
| Model weights | 2278 GB |
| KV cache | 2.9 GB |
| Linear-attention state (fixed — does not grow with context) | 0.5 GB |
| Runtime + macOS | 284 GB |
| Total used | 2565 / 16 GB |
| Short by | 2549 GB |
Max context at 8-bit: does not fit. Unified memory is shared by the OS — FitLLM leaves ~20% headroom.
Every quantization on M6 16GB
| Quant | Weights | Fits (KV F16) | Used @32K |
|---|---|---|---|
| 4bit | 1139 GB | ❌ won't fit | 1290 / 16 GB |
| 8bit | 2278 GB | ❌ won't fit | 2565 / 16 GB |
| 16bit | 4556 GB | ❌ won't fit | 5117 / 16 GB |
Lower quants free memory at some output-quality cost — 4-bit is the common sweet spot for local use.
▶ Open the interactive calculator (this exact setup)Embed this verdict
Live badge for your README or model card — recomputed by the engine, never stale:
[](https://fitllm.run/can-i-run/qwen-3-8-2-4t-a95b-on-m6-16gb)
Or from your terminal (exit 0/1 — works as a pre-download guard):
npx fitllm "Qwen 3.8 2.4T-A95B" --mac 16Why most calculators get this wrong
Qwen 3.8 2.4T-A95B is a hybrid model: only 23 of its 92 layers use full attention — the rest are linear and keep no growing KV cache. Naive calculators count every layer at full context and badly over-estimate.
Other options
same Mac Models that fit in 16GB: MiniCPM5-1B, Qwen3-0.6B, Qwen3-1.7B, Llama-3.2-1B-Instruct, Gemma-3-1B-it.
Reproduce it
Open math: fitllm-engine (MIT), from official config.json.
All numbers are computed by the open-source fitllm-engine (MIT) from official model config.json values — reproduce or audit them yourself. Estimates; real usage varies with runtime (llama.cpp / MLX / Ollama), driver and display. Found a mismatch? Report it. · FitLLM home