// Qwen · Vision-language
Qwen3-VL-235B-A22B-Thinking: VRAM & local deployment
235.7B total · 22B active (MoE) parameters. About 159 GB at 4-bit and 521 GB at BF16 with one 8K-token request.
released 2025-09-22licence: Commercial OKmemory figures are estimates
01Specs
| Publisher | Alibaba Qwen · Qwen |
|---|---|
| Family / series | Qwen · Qwen3-VL-Thinking |
| Type | Vision-language (image-text-to-text) |
| Parameters | 235.7B total · 22B active (MoE) Source: counted from official safetensors weights |
| Context length | 256K tokens (262,144)max_position_embeddings in config.json |
| Architecture | 94 layers · GQA · MoE |
| Released weights | BF16 · 471 GBdownload |
| Licence | Apache-2.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed. |
| Released | 2025-09-22 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.
| Precision | Weights | Total @ 8K | Total @ 32K | Total @ 128K | Smallest GPU | Mac |
|---|---|---|---|---|---|---|
| BF16 / FP16 (original) | 471 GB | 521 GB | 525 GB | 544 GB | 8 × H100 (640GB) | — |
| FP8 / INT8 / Q8_0 | 250 GB | 278 GB | 282 GB | 301 GB | 4 × RTX PRO 6000 (384GB) | Mac 512GB |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 143 GB | 159 GB | 164 GB | 183 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB |
KV cache per request: 8K 1.6 GB · 32K 6.3 GB · 128K 25.2 GB
Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →
03How to run it
Only runtimes the publisher or the runtime’s official library names for this model.
- Transformers — mentioned in the official model card
- llama.cpp — official GGUF weights
04Official repositories & quantised builds
3 official repo(s) from Qwen.
05Other Qwen models
All 112 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| Qwen3-VL-2B-Thinking · same series | 2.13B | 2.9 GB | 2025-10-19 |
| Qwen3-VL-4B-Thinking · same series | 4.44B | 4.7 GB | 2025-10-11 |
| Qwen3-VL-8B-Thinking · same series | 8.77B | 7.6 GB | 2025-10-11 |
| Qwen3-VL-30B-A3B-Thinking · same series | 31.1B / 3B | 22.0 GB | 2025-09-30 |
| Qwen3-VL-32B-Thinking · same series | 33.4B | 24.9 GB | 2025-10-19 |
| Qwen-Image-2.1-Turbo | 7.12B | 5.2 GB | 2026-10-09 |
| Qwen-Image-2.1 | 7.12B | 5.2 GB | 2026-09-14 |
| Qwen3.8-Flash-Next | 180B / 6B | 121 GB | 2026-08-24 |
| Qwen3.8-2.4T-A95B | 2.4T / 95B | 1.60 TB | 2026-08-08 |
| Qwen3.8-27B | 27.8B | 19.6 GB | 2026-08-05 |
| Qwen3.6-27B | 27.8B | 19.6 GB | 2026-04-21 |
| Qwen3.6-35B-A3B | 36B / 3B | 24.6 GB | 2026-04-15 |
| Qwen3.5-0.8B | 873M | 1.2 GB | 2026-02-28 |
06Sources
Data checked 2025-11-26. Parameters, context and licence are copied from the official repository; memory figures are our estimates.