Qwen 3 Coder 30B
#25 v Open-source jazykové modelyalibaba · v3 · coder 30b · od 2025-07-31 · 3× · naposledy 02. 7. 2026
26
Momentum
Qwen 3 Coder 30B is a specialized code model from Alibaba. According to the excerpts, it serves as a dedicated code specialist with a price of $0.35 per million tokens.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Inference Speed | Approx. 101.8 tokens/sec (Alibaba API, measured by Artificial Analysis; median of comparable models: 98.3 t/s); TTFT: 2.73 s (Alibaba API) |
| Context Window | 262,144 tokens native (per official Hugging Face model card); expandable up to 1M tokens via YaRN |
| Model Size (Parameters) | 30.5B total parameters (MoE); of which 3.3B active per inference forward pass (128 experts, 8 active) |
| Memory Requirement | Approx. 21.9 GB VRAM at Q4_K_M quantization (recommended: ≥26 GB VRAM); approx. 67 GB VRAM at FP16 full density; min. 18.6 GB RAM for GGUF Q4_K_M |