Contributed byMario AlkaQwen (Alibaba) Qwen3-Coder-Next
Performance benchmark · measured on 29.07.2026 03:01
Benchmark-ID
run-20260729-032132-1f4b4dTimebench 3 - Kombi (Prefill + Generation)MoERuntime: llama.cppQuantisierung: Q4_K_M
Generation107,02tok/s
Prefill457,28tok/s
Time to First Token125.536,50ms
Total duration665,73s
Concurrency10parallel
Ranking in the field
22of 29 systems
Performance benchmark · Primary metric: Generation-Speed (tok/s) · 10× concurrent
This run is better than 25 % of all comparable systems.
Generation
107,0 tok/s
-89 % vs Ø 960,8
Prefill
457,3 tok/s
-96 % vs Ø 10.710,5
Time to First Token
125.537 ms
+225 % vs Ø 38.678
Distribution in the field1 – 2.491 tok/s
Wie schlägt sich dieser Benchmark mit anderen Modellen?
gemma-4-E2B-itNVIDIA GeForce RTX 5090 · run-20260728-184455-e28775
2.491,2 tok/s
gpt-oss-20bNVIDIA GeForce RTX 5090 · run-20260729-032121-7c961d
1.890,7 tok/s
NVIDIA-Nemotron-3-Nano-4BNVIDIA GeForce RTX 5090 · run-20260728-194135-2ed9dd
1.878,7 tok/s
Laguna-XS-2.1NVIDIA GeForce RTX 5090 · run-20260729-032123-a1af22
1.765,0 tok/s
Laguna-XS-2.1NVIDIA GeForce RTX 5090 · run-20260729-032123-1156eb
1.740,1 tok/s
Laguna-XS-2.1NVIDIA GeForce RTX 5090 · run-20260729-032122-887a06
1.738,4 tok/s
gemma-4-E4B-itNVIDIA GeForce RTX 5090 · run-20260728-194134-ce100b
1.606,1 tok/s
Meta-Llama-3.1-8B-InstructNVIDIA GeForce RTX 5090 · run-20260729-060803-89a23b
1.593,1 tok/s
Mamba-Codestral-7B-v0.1NVIDIA GeForce RTX 5090 · run-20260729-060802-bea1f6
1.571,0 tok/s
North-Mini-Code-1.0NVIDIA GeForce RTX 5090 · run-20260729-032131-92781e
1.557,3 tok/s
gpt-oss-20bNVIDIA GeForce RTX 5090 · run-20260724-004604-1c9023
1.386,8 tok/s
gpt-oss-20bNVIDIA GeForce RTX 5090 · run-20260723-045447-edeed3
1.295,3 tok/s
gemma-4-12B-itNVIDIA GeForce RTX 5090 · run-20260729-060805-568759
1.058,9 tok/s
Qwen3-Coder-30B-A3B-InstructNVIDIA GeForce RTX 5090 · run-20260723-045450-366dae
966,2 tok/s
Qwen3-Coder-Next this runNVIDIA GeForce RTX 5090 · run-20260729-032132-1f4b4d
107,0 tok/s
How does this benchmark compare on other GPUs?
Same model on different hardware · 10× concurrent · Generation (tok/s)
NVIDIA RTX PRO 6000 Blackwell Workstation Editionrun-20260728-184455-9c52c0
1.253,3 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260728-184455-d2b6cf
1.147,5 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032132-7b9d5e
1.142,0 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032132-c92e3e
1.133,5 tok/s
3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Editionrun-20260729-032131-4fa06a
1.089,3 tok/s
3× AMD Radeon AI PRO R9700run-20260723-032012-f81bf7
156,4 tok/s
NVIDIA GeForce RTX 5090 this runrun-20260729-032132-1f4b4d
107,0 tok/s
NVIDIA GeForce RTX 5070 Tirun-20260728-184455-7fc929
106,1 tok/s
NVIDIA GeForce RTX 3090 Tirun-20260729-032119-98f2f8
78,2 tok/s
Hardware
GPU: NVIDIA GeForce RTX 5090 · 32 GB VRAM
CPU: AMD Ryzen 7 5800X3D 8-Core Processor
RAM: 126 GB
Mainboard: ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING
Setup
Runtime: llama.cpp
Quantization: Q4_K_M
Model: Qwen3-Coder-Next
Configuration
# LLM-Benchmark Konfiguration
# Modell : Qwen3-Coder-Next
# Engine : llama.cpp
# Run-ID : run-20260729-032132-1f4b4d
# GPU : NVIDIA GeForce RTX 5090
# CPU : AMD Ryzen 7 5800X3D 8-Core Processor
# RAM : 126 GB
bench@llm-benchmark:~$ /root/llama.cpp/build/bin/llama-server \
-m /root/.cache/huggingface/hub/models--unsloth--Qwen3-Coder-Next-GGUF/snapshots/ce09c67b53bc8739eef83fe67b2f5d293c270632/Qwen3-Coder-Next-Q4_K_M.gguf \
--alias Qwen3-Coder-Next \
--host 0.0.0.0 \
--port 8000 \
-ngl 30 \
-c 16384 \
-np 4 \
--jinja
| Engine?Die Inferenz-Software, die das Modell ausliefert (z.B. vLLM oder llama.cpp). Sie bestimmt Geschwindigkeit, unterstuetzte Modellformate und welche Parameter ueberhaupt verfuegbar sind. | llamacpp |
| Modellalias?Der Name, unter dem das Modell ueber die API angesprochen wird. Genau dieser Wert muss im Request-Feld 'model' stehen. | Qwen3-Coder-Next |
| Kontextlaenge?Maximale Anzahl Tokens (Eingabe + erzeugte Ausgabe zusammen), die das Modell pro Anfrage verarbeiten kann. | 16384 |
| Modellpfad?Pfad zur GGUF-Modelldatei, die geladen und ausgeliefert wird. | /root/.cache/huggingface/hub/models--unsloth--Qwen3-Coder-Next-GGUF/snapshots/ce09c67b53bc8739eef83fe67b2f5d293c270632/Qwen3-Coder-Next-Q4_K_M.gguf |
| GPU-Layer?Anzahl der auf die GPU ausgelagerten Modell-Layer. Hoeher = mehr VRAM und schneller; der Rest laeuft auf der CPU. 999 = alles auf GPU. | 30 |
| Kontext?Groesse des Kontextfensters in Token. 0 = der beim Training verwendete Kontext des Modells. | 16384 |
| np | 4 |
Anzeige
Model comparison
Qwen3-Coder-Next on various hardware
All published performance runs of this model – each bubble a variant: position = prefill (X) × generation (Y), bubble size = number of runs. Closer to the top right = faster. ★ Marked gold = this benchmark.
GPUby graphics card
CPUby processor
AMD Ryzen 9 9950X 16-Core Processor 1.253,3 tok/sAMD Ryzen Threadripper PRO 9965WX 24-Cores 1.147,5 tok/sAMD Ryzen Threadripper PRO 7955WX 16-Cores 156,4 tok/s★ AMD Ryzen 7 5800X3D 8-Core Processor 107,0 tok/s this runAMD Ryzen Threadripper PRO 5975WX 32-Cores 106,1 tok/sAMD Ryzen 9 8945HX with Radeon Graphics 78,2 tok/s
MBby mainboard
ASUSTeK COMPUTER INC. ProArt X870E-CREATOR WIFI 1.253,3 tok/sASUSTeK COMPUTER INC. Pro WS WRX90E-SAGE SE 1.147,5 tok/s★ ASUSTeK COMPUTER INC. ROG STRIX B550-A GAMING 107,0 tok/s this runASUSTeK COMPUTER INC. Pro WS WRX80E-SAGE SE WIFI 106,1 tok/sMeigao Innovation Technology (Shen Zhen) Co., Ltd DRFXL (MotherBoard Series) 78,2 tok/s
DRVby driver
unbekannt 1.253,3 tok/s
💰 Economics
Economics of this run
Operating cost, TCO and comparison with the next-best runs of the same model at identical concurrency (10× concurrent). Methodology →
⚙️ ConfigurationAll metrics and charts below follow these settings – based on a 24-month runtime.Save to URLReset
⚡ Electricity price
EUR/kWh
⚙️ System utilization
100 %
🖥️ Acquisition
EUR
🔌 Idle
72 W
⚡ TDP
622 W
☁️ External LLM (API)
Electricity0.30 EUR/kWh
Avg power (incl. idle)622 W estimated (TDP)GPU 575 + CPU 35 + Board 12 W full load
Avg cost / hourEUR 0.19
Electricity / 1M tokensEUR 0.48
Token / kWh619.91K
Acquisition (system)EUR 4,986 full priceGPU EUR 3,300 · CPU EUR 349 · Board EUR 149 · RAM EUR 1,008 · PSU EUR 180
Electricity (2 years)–
TCO (2 years)EUR 8,253
Output tokens (2 years)6.75B
☁️ External LLM (API) – comparison
External LLM cost (2 years)–
Savings vs. external (2 years)–
All values above and the charts below take the configured system utilization into account: at X% the system generates only X% of the time, the rest it idles (72 W). Cost per hour drops (more idle), cost per token rises.
Cost over 2 years – electricity only
Cost over 2 years – incl. acquisition (TCO)
Speed vs. tokens per euro
Euro per 1M tokens
Comparison vs. API – economics per benchmark
| Qwen3-Coder-NextNVIDIA GeForce RTX 5090 | Qwen3-Coder-NextNVIDIA RTX PRO 6000 Blackwell Workstation Edition | Qwen3-Coder-Next3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | Qwen3-Coder-Next3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | |
|---|---|---|---|---|
| Electricity cost (24 mo.) | – | – | – | – |
| Acquisition cost | – | – | – | – |
| Total cost (TCO) | – | – | – | – |
| Generated tokens (24 mo.) | – | – | – | – |
| Token price via API | – | – | – | – |
| Break-even point (days) | – | – | – | – |
| Result (savings / extra cost) | – | – | – | – |
NVIDIA GeForce RTX 5090 · this run 107 tok/sNVIDIA RTX PRO 6000 Blackwell Workstation Edition 1,253 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1,148 tok/s3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition 1,142 tok/s
Comparison with up to 3 next-best runs of this model at the same concurrency (at least one on different hardware). Power = GPU TDP + CPU (idle + 15 %) + board (estimated), acquisition = full system (GPU + CPU + board + RAM + PSU), prices = stored market prices.
Contributed by
Mario Alka Administrator
@marioalka
Ich bin Unternehmer, Softwareentwickler und KI-Enthusiast. Seit vielen Jahren entwickle ich Unternehmenssoftware und beschäftige mich inzwischen fast täglich mit lokalen LLMs, KI-Agenten und leistungsfähiger KI-Hardware.
Mit LLM-Benchmark.de möchte ich eine Plattform schaffen, auf der Modelle, GPUs und Agenten objektiv und reproduzierbar miteinander verglichen werden.