HiRAM ▮▮

Local AI hardware guide

What machines can run Qwen locally?

Updated July 28, 2026 · HiRAM Deals

For the small stack, start with Qwen 3.5 9B on a 32 GB machine. Ollama’s current quantized package is about 6.6 GB and advertises a 256K context window, leaving the rest of system memory for the context cache, runtime, and operating system.

Our recommendation: Pair Qwen 3.5 9B with 32 GB RAM for a compact local stack. That leaves far more room for a 100K-token test than a 16 GB machine. Choose 64 GB if you use long context often.

Qwen 3.5 9B accepts text and images. Ollama’s default package is about 6.6 GB and advertises a 256K context window. It fits a 32 GB desktop more comfortably than the 27B-and-larger models.

32 GB RAM

Recommended for Qwen 3.5 9B. It leaves meaningful capacity beyond the 6.6 GB quantized weights for a large KV cache and normal system overhead.

64 GB RAM

The safer long-context workstation. Choose this when you expect 100K-token prompts frequently, keep development tools open, or want larger quantizations.

96–128 GB RAM

Useful for 27B-class models, multiple concurrent models, CPU-first deployments, or avoiding aggressive memory compromises.

Can 32 GB handle a 100K-token context?

Qwen 3.5 9B makes a 100K-token test plausible in 32 GB because the model package is relatively small. Ollama's advertised 256K limit does not guarantee that every configuration fits. Cache format, image inputs, runtime version, thinking output, and open applications all change the footprint.

Use 32 GB as the value-oriented target for a single-user 100K-token experiment. Use 64 GB when you need that context length reliably, want fewer tuning compromises, or plan to run other memory-heavy software at the same time.

Install with Ollama

Install Ollama, then run the 9B model explicitly:

ollama run qwen3.5:9b

Review the exact download size and context limit before selecting a tag in the Qwen 3.5 library on Ollama.

Hardware priorities

Primary references