Qwen3 30B-A3B
Data checked: 2026-10-12. Review interval shown below. Prices need review after 30 days; specifications and benchmark collections after 90 days. Dates describe the recorded check, not a live retailer quote.
- Verified Checked against the primary source (for example the model’s config.json on Hugging Face).
- Sourced Taken from a cited source; not yet independently checked.
- Estimate Calculated with the published formulas; not a measurement.
- Sample Placeholder data for layout; not a real figure.
- Not yet measured No public, citable score found for this benchmark yet. We never fill the gap with a guess.
Verified metadata is not a measured performance result. Only cited runs matching model, device, quant and recorded context can replace speed estimates. Estimates assume one stream and F16 KV cache, with a ±25% uncertainty band. A published run applies only to its recorded runtime and conditions; missing settings are unknown.
Evidence labels · Freshness policy · Calculation methodology
At a glance
A mixture-of-experts model that reads only 3.3B parameters per token. Needs 24 GB to hold, but decodes like a small model.
At Q4_K_M and 8k context it needs about 19.9 GB: 18.6 GB of weights, 0.8 GB of KV cache and 0.6 GB of runtime overhead. Only 3.3B of its 30.5B parameters are read per token, so it decodes far faster than its size suggests (see memory bandwidth).
Estimated setup
Quant | Q4_K_M |
|---|---|
Smallest standard card size at 8k | 24 GB |
Max context, 24 GB card | 32k |
Runtimes | llama.cpp, Ollama, LM Studio, MLX |
Quantization options
Total memory = weights + KV cache (F16) + runtime overhead. File sizes marked Verified are the sum of GGUF files from Qwen/Qwen3-30B-A3B-GGUF on the Hugging Face API. KV cache costs 0.098 MB per token before sliding-window savings.
Quant | bpw | File | At 8k | At 32k | At 128k | KL-divergence | 24 GB card, 8k |
|---|---|---|---|---|---|---|---|
| Q3_K_M | 3.91 | 14.9 GBEstimate | 16.3 GB | 18.9 GB | — | 0.071Sample | Fits |
| Q4_K_M | 4.85 | 18.6 GBVerified | 19.9 GB | 22.5 GB | — | 0.017Sample | Fits |
| Q5_K_M | 5.69 | 21.7 GBVerified | 23.1 GB | 25.7 GB | — | 0.008Sample | Tight |
| Q6_K | 6.56 | 25.1 GBVerified | 26.5 GB | 29.1 GB | — | 0.004Sample | Needs offload |
| Q8_0 | 8.50 | 32.5 GBVerified | 33.8 GB | 36.5 GB | — | 0.001Sample | Needs offload |
| F16 | 16.00 | 61.1 GBEstimate | 62.4 GB | 65.0 GB | — | reference | Needs offload |
Why this matters: What is quantization? · What is a KV cache?
Decode speed by device
Q4_K_M, single stream, 8k context. Estimates use the brief §7.3 formula (η × bandwidth ÷ bytes per token) and are ±25% until measured.
Speed table
Device | Verdict | Decode, tok/s | Range | Runtime |
|---|---|---|---|---|
| GeForce RTX 5090 32 GB | Fits | 382.4Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 4090 24 GB | Fits | 215.1Estimate | ±25% | llama.cpp (estimate) |
| Radeon RX 7900 XTX 24 GB | Fits | 204.9Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 3090 24 GB (used) | Fits | 199.8Estimate | ±25% | llama.cpp (estimate) |
| 2× RTX 3090 (used) 48 GB | Fits | 199.8Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Ultra (base) | Fits | 192.1Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Ultra · 512 GB | Fits | 192.1Estimate | ±25% | llama.cpp (estimate) |
| MacBook Pro M4 Max 40c · 128 GB | Fits | 87.4Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Max (base) | Fits | 73.6Estimate | ±25% | llama.cpp (estimate) |
| NVIDIA DGX Spark FE 128 GB | Fits | 43.7Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 5060 Ti 16 GB | Needs offload | 42.3Estimate | ±25% | llama.cpp (estimate) |
| Ryzen AI Max+ 395 mini PC 128 GB | Fits | 41.0Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 4060 Ti 16 GB | Needs offload | 35.1Estimate | ±25% | llama.cpp (estimate) |
| CPU only · Ryzen 9 9950X + 128 GB DDR5 | Fits | 15.9Estimate | ±25% | llama.cpp (estimate) |
Compare with similar-sized models
Model | Parameters | Quality score | Licence |
|---|---|---|---|
| Qwen3 30B-A3B | 30.5 B | 1,317Sourced | Apache-2.0 |
| gpt-oss-20b | 20.9 B | 1,287Sourced | Apache-2.0 |
| Mistral Small 3.2 24B | 24.0 B | 1,338Sourced | Apache-2.0 |
LMArena text, overall, entry 'qwen3-30b-a3b': 1316.8 (95% CI 1312.1–1321.5, 26,160 votes), leaderboard published 2026-10-08. CC-BY-4.0. Dataset
llama-perplexity --kl-divergence on a fixed corpus. Speed and quality are never merged into one score. Sample values are illustrative and must not be used to select a model.Architecture
Parameters, total | 30.53 BVerifiedsafetensors.total from the HF API. Source |
|---|---|
Parameters, active per token | 3.30 BSourcedActive parameters per token from the model card. Source |
Layers | 48VerifiedSource |
Query heads | 32VerifiedSource |
KV heads (GQA) | 4VerifiedSource |
Head dimension | 128VerifiedSource |
Hidden size | 2048VerifiedSource |
Experts | 128 experts, 8 active per tokenVerifiedSource |
Operating conditions
Max context | 32,768 tokensSourcedconfig.json lists max_position_embeddings 40,960; the model card gives 32,768 native and 131,072 with YaRN. Source |
|---|---|
Licence | Apache-2.0VerifiedSource |
Weights | Qwen/Qwen3-30B-A3B |
GGUF files | Qwen/Qwen3-30B-A3B-GGUF |
Usable memory, discrete | VRAM − 0.5 GB2 |
Usable memory, unified | RAM × 0.752 |
Fits when | total ≤ 90% of usable2 |
1 Estimates: decode ≈ η × bandwidth ÷ (active weights + KV cache), η 0.60 discrete / 0.45 unified / 0.50 CPU (concept brief §7.3). Not measurements.
2 Fit rules from the concept brief §7.3, including the 0.75 unified-memory factor used by canirun.ai.
Flags: Verified = checked against config.json / HF API; Sourced = from a cited source; Estimate = formula; Sample = placeholder. Markdown · JSON · Report a problem