Qwen 3.5 32B Coder
The gold-standard single-GPU open-weight coding model: 32.5B parameters trained on 12 trillion code and repository commit tokens with native IDE Fill-in-the-Middle support.
Provider
Alibaba Qwen (Apache 2.0)
Architecture
Dense Transformer (32.5B) + GQA + Fill-in-the-Middle (FIM)
Parameters
32.5B Dense
Context Window
131,072 tokens (Max Output: 32,768 tokens)
API Pricing
Input: $0.14 / 1M · Cached: $0.05 / 1M · Output: $0.30 / 1M
VRAM (16K Ctx)
Q4_K_M: 22.46 GB · Q8_0: 37.6 GB · FP16: 69.88 GB
Hardware Target
NVIDIA RTX 4090 (24GB) or RTX 5090 (32GB); fits 16GB at IQ3_M
Interactive Token Cost Estimator (Qwen 3.5 32B Coder)
GPU Hardware Compatibility Matrix (Qwen 3.5 32B Coder at 16K Context)
| GPU / Hardware Profile | VRAM | Q4_K_M (4.8-bit) | Q8_0 (8-bit) | Est. Speed (Q4) |
|---|---|---|---|---|
| RTX 5060 Ti 16GB 448 GB/s | 16 GB | REQUIRES MULTI-GPU (22.46 GB) | REQUIRES MULTI-GPU (37.6 GB) | ~3 tok/s |
| RTX 4090 24GB 1008 GB/s | 24 GB | TIGHT FIT (Q8 KV) (22.46 GB) | REQUIRES MULTI-GPU (37.6 GB) | ~24 tok/s |
| RTX 5090 32GB 1792 GB/s | 32 GB | FITS IN VRAM (22.46 GB) | REQUIRES MULTI-GPU (37.6 GB) | ~59 tok/s |
| M4 Max 64GB 546 GB/s | 52 GB | FITS IN VRAM (22.46 GB) | FITS IN VRAM (37.6 GB) | ~18 tok/s |
| 2× 24GB (48GB) 920 GB/s | 48 GB | FITS IN VRAM (22.46 GB) | FITS IN VRAM (37.6 GB) | ~30 tok/s |
| H100 80GB 3350 GB/s | 80 GB | FITS IN VRAM (22.46 GB) | FITS IN VRAM (37.6 GB) | ~111 tok/s |
| M4 Ultra 192GB 819 GB/s | 164 GB | FITS IN VRAM (22.46 GB) | FITS IN VRAM (37.6 GB) | ~27 tok/s |
| 8× H200 (1,128GB) 4800 GB/s | 1128 GB | FITS IN VRAM (22.46 GB) | FITS IN VRAM (37.6 GB) | ~159 tok/s |
Architectural Strengths & Target Workloads
- Fits natively on a single 24GB GPU (RTX 4090/3090) at Q4_K_M (~19.5 GB weights)
- Can run on 16GB GPUs (RTX 5060 Ti 16GB) using IQ3_M quantization (~14.0 GB weights)
- Matches proprietary mid-tier APIs across TypeScript, Rust, Go, Python, and Kubernetes manifests
Primary Engineering Workloads: Local IDE Autocomplete & Agent · Air-Gapped Code Review · Single-GPU Dev Server.
Frequently Asked Questions (Qwen 3.5 32B Coder)
How much does Qwen 3.5 32B Coder cost per 1M tokens?
Qwen 3.5 32B Coder is priced at $0.14 per 1M uncached input tokens, $0.05 per 1M cached input tokens, and $0.30 per 1M output tokens (Local Single-GPU (RTX 4090 24GB / RTX 5090 32GB) or OpenRouter API).
What is the context window of Qwen 3.5 32B Coder?
It supports up to 131,072 tokens (131,072 tokens) and a maximum generation output of 32,768 tokens.