Qwen3 32B
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
The largest dense Qwen3. Q4_K_M weights are 19.8 GB, which is tight on a 24 GB card once context grows.
At Q4_K_M and 8k context it needs about 22.5 GB: 19.8 GB of weights, 2.1 GB of KV cache and 0.6 GB of runtime overhead. It is dense, so every token reads all 19.8 GB of weights; speed scales with memory bandwidth.
Estimated setup
Quant | Q4_K_M |
|---|---|
Smallest standard card size at 8k | 32 GB |
Max context, 24 GB card | 8k |
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/Qwen3-32B-GGUF on the Hugging Face API. KV cache costs 0.262 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 | 16.0 GBEstimate | 18.7 GB | 25.3 GB | — | 0.071Sample | Fits |
| Q4_K_M | 4.85 | 19.8 GBVerified | 22.5 GB | 29.1 GB | — | 0.017Sample | Tight |
| Q5_K_M | 5.69 | 23.2 GBVerified | 25.9 GB | 32.5 GB | — | 0.008Sample | Needs offload |
| Q6_K | 6.56 | 26.9 GBVerified | 29.6 GB | 36.2 GB | — | 0.004Sample | Needs offload |
| Q8_0 | 8.50 | 34.8 GBVerified | 37.5 GB | 44.2 GB | — | 0.001Sample | Needs offload |
| F16 | 16.00 | 65.5 GBEstimate | 68.2 GB | 74.9 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 | 49.1Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 4090 24 GB | Tight | 27.6Estimate | ±25% | llama.cpp (estimate) |
| Radeon RX 7900 XTX 24 GB | Tight | 26.3Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 3090 24 GB (used) | Tight | 25.6Estimate | ±25% | llama.cpp (estimate) |
| 2× RTX 3090 (used) 48 GB | Fits | 25.6Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Ultra (base) | Fits | 24.6Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Ultra · 512 GB | Fits | 24.6Estimate | ±25% | llama.cpp (estimate) |
| MacBook Pro M4 Max 40c · 128 GB | Fits | 11.2Estimate | ±25% | llama.cpp (estimate) |
| Mac Studio M5 Max (base) | Fits | 9.4Estimate | ±25% | llama.cpp (estimate) |
| NVIDIA DGX Spark FE 128 GB | Fits | 5.6Estimate | ±25% | llama.cpp (estimate) |
| Ryzen AI Max+ 395 mini PC 128 GB | Fits | 5.3Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 5060 Ti 16 GB | Needs offload | 3.7Estimate | ±25% | llama.cpp (estimate) |
| GeForce RTX 4060 Ti 16 GB | Needs offload | 3.3Estimate | ±25% | llama.cpp (estimate) |
| CPU only · Ryzen 9 9950X + 128 GB DDR5 | Fits | 2.0Estimate | ±25% | llama.cpp (estimate) |
Compare with similar-sized models
Model | Parameters | Quality score | Licence |
|---|---|---|---|
| Qwen3 32B | 32.8 B | 1,340Sourced | 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 'qwen3-32b': 1340.1 (95% CI 1330.8–1349.4, 3,926 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 | 32.76 BVerifiedsafetensors.total from the HF API. Source |
|---|---|
Parameters, active per token | 32.76 BVerifiedDense: all parameters active. Source |
Layers | 64VerifiedSource |
Query heads | 64VerifiedSource |
KV heads (GQA) | 8VerifiedSource |
Head dimension | 128VerifiedSource |
Hidden size | 5120VerifiedSource |
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/Qwen3-32B |
GGUF files | Qwen/Qwen3-32B-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