Qwen3-Coder 30B-A3B
The workhorse local coding model: 3.3B active parameters means it runs fast on almost anything with 24 GB, including CPU-only machines. Quality is well below the large MoEs, but the latency is what makes an autocomplete loop usable.
Alibaba QwenMixture of expertscoding26 machines can run it
Total parameters
30.532B
Official spec sets memory need
Active parameters
3.3B
Official spec sets decode speed
Context
256K
Official spec tokens
4-bit weights
17 GB
Calculated before KV cache
This is a sparse mixture of experts. All 30.532B of weights must be available to the runtime, but only 3.3B are read per generated token. Ordinary runtimes keep the full quantized model in memory; a runtime-specific SSD-streaming artifact can retain a smaller working set and fetch expert data on demand, trading speed for capacity.
Model specification
Publisher
Alibaba Qwen Official spec
Architecture
Sparse mixture of experts Official spec
Total parameters
30.532B Official spec
Active parameters
3.3B per token Official spec
Context length
262,144 tokens Official spec
Attention ⓘ
GQA · 4 KV heads × 128 dim × 48 layers Official spec
Licence
Apache 2.0 Official spec source
Specification confidence ⓘ
High · verified 2026-10-19 Official spec source
Released
31 Jul 2025 Official spec
Official source
Quantizations and memory
| Quantization | Format | Bits/weight | Resident | Download | Quality kept |
|---|---|---|---|---|---|
MLX 4-bit | mlx | 4.5 | 16.6 GB | 16.6 GB | 98.0% |
Q4_K_Mdefault | gguf | 4.85 | 17.8 GB | 17.8 GB | 98.5% |
Q5_K_M | gguf | 5.69 | 20.8 GB | 20.8 GB | 99.3% |
Q8_0 | gguf | 8.5 | 30.8 GB | 30.8 GB | 99.9% |
BF16 | safetensors | 16 | 56.9 GB | 56.9 GB | 100.0% |
Generic weight sizes are computed from the parameter count and bits per weight plus a format-specific overhead. Runtime-specific artifacts use their published resident and download footprints; streamed models can therefore require much more disk than memory. Quality retention is an assumption, not a measured evaluation.
Coding & quality benchmarksCompare coding results →
No published benchmark results have been imported for this model yet. This is missing evidence, not a score of zero.
We show source metrics rather than deriving one opaque quality number. Different benchmarks measure genuinely different things, and collapsing them into a single score would hide exactly the disagreements worth seeing.
Recommendations
Cheapest that can run it
~8.8–12.7 t/s · $1,531
Lowest purchase price among configurations where the model fits at some quantization in our catalogue. Speed is not considered.
Cheapest above 20 t/s
~20.5–29.5 t/s · $1,711
20 tokens/second is roughly the point at which generation keeps pace with reading. Below it, interactive use feels like waiting.
Cheapest above 40 t/s
~197–284 t/s · $1,981
40 tokens/second is the threshold most people describe as comfortable for coding agents, where output arrives faster than you can review it.
Fastest with real measurements
52 t/s measured at Q4_K_M · $2,161
Highest throughput among configurations with an actual published measurement rather than our estimate.
Best throughput per purchase-price unit
~221–317 t/s · $2,071
Highest decode tokens/second per 1,000 units of the displayed purchase currency. Ignores running costs and resale — see the economics section for the full picture.
Hardware that runs Qwen3-Coder 30B-A3B1 measured, 25 estimated
26 of 26 rows
| Hardware↕ | Quant | Memory↕ | Decode▼ | Prefill↕ | Context | Price↕ | Fit | Confidence↕ |
|---|---|---|---|---|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) NVIDIA · 1,128 GB · 38,400 GB/s | Q8_0recommended | 36.3 GB | ~529–761 t/s | ~266560–553620 t/s | 256K | $359,430 | Comfortable | Estimated |
| Quad RTX 5090 workstation (4x 32 GB) NVIDIA · 128 GB · 7,168 GB/s | Q8_0recommended | 34.7 GB | ~301–434 t/s | ~9980–20730 t/s | 256K | $12,611 | Comfortable | Estimated |
| RTX 5090 workstation (1x 32 GB) NVIDIA · 32 GB · 1,792 GB/s | Q5_K_Mrecommended | 23.2 GB | ~299–430 t/s | ~3070–6380 t/s | 64K | $3,332 | Comfortable | Estimated |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) Lenovo · 384 GB · 7,168 GB/s | Q8_0recommended | 34.7 GB | ~279–402 t/s | ~12800–26580 t/s | 256K | $58,553 | Comfortable | Estimated |
| Dual RTX 5090 workstation (2x 32 GB) NVIDIA · 64 GB · 3,584 GB/s | Q8_0recommended | 33.9 GB | ~236–339 t/s | ~4990–10370 t/s | 192K | $5,854 | Comfortable | Estimated |
| Dell Precision 7960 Rack (2x RTX PRO 6000, 192 GB) Dell · 192 GB · 3,584 GB/s | Q8_0recommended | 33.9 GB | ~223–320 t/s | ~6400–13290 t/s | 256K | $36,663 | Comfortable | Estimated |
| Used RTX 4090 workstation (24 GB) NVIDIA · 24 GB · 1,008 GB/s | Q4_K_Mrecommended | 20.0 GB | ~221–317 t/s | ~3420–7100 t/s | 16K | $2,071 | Borderline | Estimated |
| RTX PRO 6000 Blackwell workstation (96 GB) NVIDIA · 96 GB · 1,792 GB/s | Q8_0recommended | 33.5 GB | ~210–303 t/s | ~3240–6730 t/s | 256K | $10,359 | Comfortable | Estimated |
| RTX PRO 6000 Max-Q workstation (96 GB, 300 W) NVIDIA · 96 GB · 1,792 GB/s | Q8_0recommended | 33.5 GB | ~210–303 t/s | ~2650–5510 t/s | 256K | $10,809 | Comfortable | Estimated |
| Dual used RTX 4090 workstation (48 GB) NVIDIA · 48 GB · 2,016 GB/s | Q5_K_Mrecommended | 23.6 GB | ~210–302 t/s | ~5560–11540 t/s | 192K | $3,692 | Comfortable | Estimated |
| RTX 3090 workstation (Ryzen 9 7950X, 64 GB) NVIDIA · 24 GB · 936 GB/s | Q4_K_Mrecommended | 20.0 GB | ~197–284 t/s | ~1470–3060 t/s | 16K | $1,981 | Borderline | Estimated |
| NVIDIA DGX Station GB300 (748 GB) NVIDIA · 748 GB · 7,100 GB/s | Q8_0recommended | 33.5 GB | ~168–242 t/s | ~35620–73970 t/s | 256K | $90,082 | Comfortable | Estimated |
| Mac Studio M5 Ultra 512 GB Apple · 512 GB · 1,200 GB/s | Q8_0recommended | 33.5 GB | ~167–240 t/s | ~2890–6010 t/s | 256K | $11,350 | Comfortable | Estimated |
| Mac Studio M5 Ultra 256 GB Apple · 256 GB · 1,200 GB/s | Q8_0recommended | 33.5 GB | ~167–240 t/s | ~2890–6010 t/s | 256K | $8,647 | Comfortable | Estimated |
| Mac Studio M5 Ultra 96 GB Apple · 96 GB · 1,200 GB/s | Q8_0recommended | 33.5 GB | ~167–240 t/s | ~2310–4800 t/s | 256K | $5,945 | Comfortable | Estimated |
| Radeon AI PRO R9700 workstation (32 GB) AMD · 32 GB · 644 GB/s | Q5_K_Mrecommended | 23.2 GB | ~154–222 t/s | ~2450–5090 t/s | 64K | $2,521 | Comfortable | Estimated |
| Mac Studio M5 Max 36 GB Apple · 36 GB · 614 GB/s | Q5_K_Mrecommended | 23.2 GB | ~134–192 t/s | ~1160–2400 t/s | 64K | $2,702 | Comfortable | Estimated |
| Mac Studio M3 Ultra 512 GB Apple · 512 GB · 819 GB/s | Q8_0recommended | 33.5 GB | ~101–145 t/s | ~697–1450 t/s | 256K | $8,287 | Comfortable | Estimated |
| Mac Studio M3 Ultra 256 GB Apple · 256 GB · 819 GB/s | Q8_0recommended | 33.5 GB | ~101–145 t/s | ~697–1450 t/s | 256K | $5,674 | Comfortable | Estimated |
| Mac Studio M5 Max 128 GB Apple · 128 GB · 614 GB/s | Q8_0recommended | 33.5 GB | ~94.6–136 t/s | ~1450–3000 t/s | 256K | $4,323 | Comfortable | Estimated |
| Mac Studio M5 Max 64 GB Apple · 64 GB · 614 GB/s | Q8_0recommended | 33.5 GB | ~94.6–136 t/s | ~1450–3000 t/s | 192K | $3,512 | Comfortable | Estimated |
| Mac mini M4 Pro 64 GB Apple · 64 GB · 273 GB/s | Q8_0recommended | 33.5 GB | ~36.3–52.2 t/s | ~121–252 t/s | 192K | $2,251 | Comfortable | Estimated |
| NVIDIA DGX Spark 128 GB NVIDIA · 128 GB · 273 GB/s | Q8_0recommended | 33.5 GB | ~34.1–49.1 t/s | ~1240–2580 t/s | 256K | $3,873 | Comfortable | Estimated |
| Framework Desktop (Ryzen AI Max+ 395, 128 GB) AMD · 128 GB · 256 GB/s | Q8_0recommended measured: Q4_K_M | 33.5 GB | ~23.5–33.8 t/s Measured 52 t/s | ~344–714 t/s | 256K evidence 4K | $2,161 | Comfortable | Estimated Measured · llama.cpp |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB) AMD · 128 GB · 256 GB/s | Q8_0recommended | 33.5 GB | ~20.5–29.5 t/s | ~344–714 t/s | 256K | $1,711 | Comfortable | Estimated |
| CPU-only workstation (Ryzen 9950X, 192 GB DDR5) Generic · 192 GB · 90 GB/s | Q8_0recommended | 33.5 GB | ~8.8–12.7 t/s | ~41.4–86.1 t/s | 256K | $1,531 | Comfortable | Estimated |
Submit a benchmarkContributions are reviewed before publication.
Similar models
Qwen2.5 Coder 32B
Alibaba Qwen · 32.764B · Apache 2.0
Qwen3.6 27B
Alibaba Qwen · 27.781B · Apache 2.0
Qwen3 30B-A3B
Alibaba Qwen · 30.532B (3.3B active) · Apache 2.0
DeepSeek Coder 33B
DeepSeek · 33.343B · DeepSeek
DeepSeek Coder V2 Lite 16B-A2.4B
DeepSeek · 15.706B (2.4B active) · DeepSeek
Qwen2.5 Coder 14B
Alibaba Qwen · 14.77B · Apache 2.0