Qwen3.5-4B on local hardware
Estimated comfort for repository coding with Aider on a MacBook Pro M4 Pro 48GB.
Qwen3.5-4BOpen source
Qwen · apache-2.0 · 7,849,305 downloads
VisionReasoning / thinking mode
- Parameters
- 4.7B exact
- Context window
- 256K tokens
- Attention
- Hybrid linear + full attention
- Layers · KV heads · head dim
- 32 · 4 · 256
- KV cache
- 32 KB/token · 1.0 GB at 32K
- Tool calling
- good (from chat template)
- Thinking
- good (from chat template)
- Vision
- good (from config)
- Commercial use
- allowedRetain the required license and attribution notices
- Import confidence
- medium · capabilities estimated
Weights by quantization
Q3_K_M
2.1 GB
gguf file size
Q4_K_M
2.6 GB
gguf file size
Q5_K_M
2.9 GB
gguf file size
Q6_K
3.3 GB
gguf file size
Q8_0
4.2 GB
gguf file size
F16 / BF16
7.8 GB
gguf file size
- • vision: good, inferred from config (low confidence).
- • Capability ratings are estimated from model size and name — this model hasn't been assessed by hand.
- • GGUF sizes come from community conversions matched by model name; equivalence and runtime support are not verified. MLX conversion availability is unverified.
Qwen3.5-4B — 6-bit
Borderline for your workload
Speed
40–50 tok/s
Excellent
Memory left
31 GB
of 48 GB
First word in
≈25 s
Comfortable
All the details
- Your setup
- MacBook Pro M4 Pro 48 GB
- Workload
- Repository coding with Aider
- Runtime → tool
- LM Studio → Aider
- Memory used
- 5.6 GB
- Context
- 32K · up to 128K practical
- Tool support
- Official
- Use-case fit
- Poorly suited
Why?
- Model fits with approximately 31 GB of memory headroom.
- LM Studio supports hardware acceleration on this machine (Metal (Apple GPU)).
- Aider can connect to LM Studio (officially supported).
- Generation (40–50 tok/s) should be appropriate for interactive coding.
- Ingesting a prompt of 14K tokens takes ≈25 s; prompt processing keeps up with interactive coding.
- 32K working context is reasonable.
- Weak coding capability for interactive coding.
Why “Borderline” rather than “Acceptable”?
- Workload suitability: Weak coding capability for interactive coding.
How sure are we?
Medium confidenceEstimated from bandwidth and compute, calibrated against 2 verified benchmarks of other models on this same chip (measured with a different engine, so calibration is partial).
- Calibrated with verified benchmarks of other models on this chip.
- Hybrid/sliding attention performance varies more between runtimes.
Medium confidence