Qwen3-Coder · released 2026-02 · 79.7B parameters (3.87B active per token)
Qwen3-Next architecture (36 Gated DeltaNet + 12 gated attention layers). Recurrent state assumed BF16.
Source configMemory for one conversation's context, on top of the weights.
| Context | FP16 | FP8 |
|---|---|---|
| 4,096 tokens | 0.13 GB | 0.08 GB |
| 8,192 tokens | 0.22 GB | 0.13 GB |
| 32,768 tokens | 0.79 GB | 0.41 GB |
| 131,072 tokens | 3.04 GB | 1.54 GB |
| 262,144 tokens | 6.04 GB | 3.04 GB |
| Format | Bits | Weights |
|---|---|---|
| BF16 / FP16 | 16 | 148.4 GB |
| FP8 (W8A8) | 8 | 75.4 GB |
| INT8 (W8A8 / W8A16) | 8 | 75.4 GB |
| NVFP4 | 4.5 | 43.4 GB |
| INT4 AWQ / GPTQ | 4.25 | 41.1 GB |
| GGUF Q8_0 | 8.5 | 78.8 GB |
| GGUF Q6_K | 6.56 | 60.8 GB |
| GGUF Q5_K_M | 5.69 | 52.8 GB |
| GGUF Q4_K_M | 4.89 | 45.4 GB |
| GGUF Q3_K_M | 3.91 | 36.3 GB |
| GGUF Q2_K | 3.35 | 31.1 GB |
One server, 8K context, FP8 on data center GPUs and GGUF Q4_K_M on consumer and Apple hardware.
| GPU | 1 user | 32 users |
|---|---|---|
| B200 SXM (HGX B200)FP8 (W8A8) | 1× · 434 tok/s | 1× · 88.5 tok/s |
| H200 SXM 141GBFP8 (W8A8) | 1× · 343 tok/s | 1× · 55.9 tok/s |
| H100 SXM 80GBFP8 (W8A8) | 2× · 290 tok/s | 2× · 70.4 tok/s |
| Instinct MI300X 192GBFP8 (W8A8) | 1× · 360 tok/s | 1× · 61.2 tok/s |
| A100 SXM 80GBINT4 AWQ / GPTQ | 1× · 293 tok/s | 1× · 40.9 tok/s |
| L40S 48GBFP8 (W8A8) | 2× · 125 tok/s | 4× · 29.5 tok/s |
| RTX PRO 6000 Blackwell Server Edition 96GBFP8 (W8A8) | 1× · 166 tok/s | 1× · 19.6 tok/s |
| GeForce RTX 5090 32GBGGUF Q4_K_M | 2× · 150 tok/s | 2× · 46.9 tok/s |
| GeForce RTX 4090 24GBGGUF Q4_K_M | 4× · 89.4 tok/s | 4× · 37.7 tok/s |
| Apple M4 Max (40-core GPU, up to 128GB)GGUF Q4_K_M | 1× · 94.2 tok/s | 1× · 11.2 tok/s |
| DGX Spark (GB10, 128GB unified)FP8 (W8A8) | 1× · 35.1 tok/s | 1× · 3.4 tok/s |