gpt-oss-20b on local hardware
Estimated comfort for repository coding with Aider on a MacBook Pro M4 Pro 48GB.
gpt-oss-20bOpen source
openai · apache-2.0 · 5,897,765 downloads
Mixture of expertsReasoning / thinking mode
- Parameters
- 21B exact
- Active per token
- ≈3.6B (4 of 32 experts)
- Context window
- 4K tokensRoPE-extended to 128K (quality degrades past native — treat long-context ratings as optimistic)
- Attention
- Sliding-window + global layers
- Layers · KV heads · head dim
- 24 · 8 · 64
- KV cache
- 24 KB/token · 0.8 GB at 32K
- Tool calling
- good (from chat template)
- Thinking
- good (from chat template)
- Vision
- basic (inferred — no supporting signal)
- Commercial use
- allowedRetain the required license and attribution notices
- Import confidence
- medium · capabilities estimated
Weights by quantization
Q3_K_M
11 GB
gguf file size
Q4_K_M
11 GB
gguf file size
Q5_K_M
11 GB
gguf file size
Q6_K
11 GB
gguf file size
MXFP4
11 GB
gguf file size
- • Context native 4096, rope-extended to 131072 (yarn); quality degrades past native — treat long-context ratings as optimistic.
- • vision: basic, inferred from missing signals (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.
gpt-oss-20b — MXFP4
Acceptable for your workload
Speed
65–85 tok/s
Excellent
Memory left
25 GB
of 48 GB
First word in
≈5 s
Excellent
All the details
- Your setup
- MacBook Pro M4 Pro 48 GB
- Workload
- Repository coding with Aider
- Runtime → tool
- LM Studio → Aider
- Memory used
- 11 GB
- Context
- 4K · up to 4K practical
- Tool support
- Official
- Use-case fit
- Poorly suited
Why?
- Model fits with approximately 25 GB of memory headroom.
- LM Studio supports hardware acceleration on this machine (Metal (Apple GPU)).
- Aider can connect to LM Studio (officially supported).
- Generation (65–85 tok/s) should be appropriate for interactive coding.
- Ingesting a prompt of 3K tokens takes ≈5 s; prompt processing keeps up with interactive coding.
- 4K window; generation drops to 65–85 tok/s when it is full. The model is limited to 4K. Aider needs at least 8K for its system prompt and tool definitions.
- Weak coding capability for interactive coding.
- gpt-oss-20b supports at most 4K context; 32K was requested.
Why “Acceptable” rather than “Comfortable”?
- Workload suitability: Weak coding capability for interactive coding.
- Context: 4K window; generation drops to 65–85 tok/s when it is full. The model is limited to 4K. Aider needs at least 8K for its system prompt and tool definitions.
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.
- MoE performance varies more between runtimes.
Medium confidence