FITMYLLM · JULY 28, 2026 · OPEN METHODOLOGY · COMMUNITY BENCHMARKS
LIVE · UPDATED DAILY
235.1B
Parameters (22B active)
Quantization Options Select your GPU for speed estimates Context length: 4K 8K 16K 32K 40K
Quant Bits VRAM @ 16K Quality IQ2_XXS 2.38 72.6 GB
70.4 + 2.2 KV
low IQ2_M 2.93 88.8 GB
86.6 + 2.2 KV
low Q2_K 3.16 95.6 GB
93.4 + 2.2 KV
low IQ3_XXS 3.25 98.2 GB
96.0 + 2.2 KV
low IQ3_XS 3.5 105.5 GB
103.3 + 2.2 KV
low Q3_K_S 3.64 109.7 GB
107.5 + 2.2 KV
low IQ3_M 3.76 113.2 GB
111.0 + 2.2 KV
low Q3_K_M 4 120.2 GB
118.0 + 2.2 KV
low Q3_K_L 4.3 129.1 GB
126.9 + 2.2 KV
moderate IQ4_XS 4.46 133.8 GB
131.6 + 2.2 KV
moderate Q4_K_S 4.67 139.9 GB
137.7 + 2.2 KV
moderate Q4_K_M 4.89 146.4 GB
144.2 + 2.2 KV
good Q5_K_S 5.57 166.4 GB
164.2 + 2.2 KV
good Q5_K_M 5.7 170.2 GB
168.0 + 2.2 KV
good Q6_K 6.56 195.5 GB
193.3 + 2.2 KV
excellent Q8_0 8.5 252.5 GB
250.3 + 2.2 KV
lossless FP16 16 472.9 GB
470.7 + 2.2 KV
lossless
Select your GPU above to see speed estimates and compatibility for each quantization.
Too big for a single GPU — plan a multi-GPU deployment
Even the lightest quant needs ~73 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →
▸ READY TO RUN THIS? RENT BY THE HOUR
RENT A GPU AND RUN QWEN3-235B-A22B NOW
Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.
Community Ratings Chat Coding Reasoning Creative Vision Roleplay Agentic
Loading ratings...
Run this model IQ2_XXS — 70.4 GB VRAM IQ2_M — 86.6 GB VRAM Q2_K — 93.4 GB VRAM IQ3_XXS — 96.0 GB VRAM IQ3_XS — 103.3 GB VRAM Q3_K_S — 107.5 GB VRAM IQ3_M — 111.0 GB VRAM Q3_K_M — 118.0 GB VRAM Q3_K_L — 126.9 GB VRAM IQ4_XS — 131.6 GB VRAM Q4_K_S — 137.7 GB VRAM Q4_K_M — 144.2 GB VRAM Q5_K_S — 164.2 GB VRAM Q5_K_M — 168.0 GB VRAM Q6_K — 193.3 GB VRAM Q8_0 — 250.3 GB VRAM FP16 — 470.7 GB VRAM
Ollama llama.cpp vLLM LM Studio KoboldCpp Jan Docker
▸ Easiest way to get started · Beginners
DOCS ↗ curl -fsSL https://ollama.com/install.sh | shCOPY
$ ollama run qwen3:235b-a22b-q4_K_MCOPY
Downloads and runs automatically. Add --verbose for speed stats.
▸ SETUP GUIDE >_
Auto-setup with fitmyllm CLI Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.
Auto-detect GPU Live tok/s in chat Speed benchmarks 9 inference engines
GPUs that can run this model At Q4_K_M quantization. Sorted by minimum VRAM.
Find the best GPU for Qwen3-235B-A22B
Build Hardware for Qwen3-235B-A22B ▸ COLOPHON FITMYLLM · INDEPENDENT · DATA-DRIVEN
FITMYLLM · EST. 2025 · © 2026
RECOMMENDATIONS FROM PUBLISHED MATH, CORRECTED BY THE COMMUNITY — 30.
▸ SPEC SHEET
Qwen3-235B-A22B — 235.1B MoE. ▸ SPECIFICATIONS
PARAMETERS 235.1B (22B active)
ARCHITECTURE Mixture of Experts
CONTEXT LENGTH 40K tokens
CAPABILITIES chat
RELEASE DATE 2025-04-28
PROVIDER Alibaba
FAMILY qwen ▸ VRAM REQUIREMENTS
QUANT BPW VRAM QUALITY IQ2_XXS 2.38 70.4 GB 65% IQ2_M 2.93 86.6 GB 75% Q2_K 3.16 93.4 GB 78% IQ3_XXS 3.25 96.0 GB 82% IQ3_XS 3.5 103.3 GB 84% Q3_K_S 3.64 107.5 GB 85% IQ3_M 3.76 111.0 GB 86% Q3_K_M 4 118.0 GB 88% Q3_K_L 4.3 126.9 GB 90% IQ4_XS 4.46 131.6 GB 92% Q4_K_S 4.67 137.7 GB 93% Q4_K_M 4.89 144.2 GB 94% Q5_K_S 5.57 164.2 GB 96% Q5_K_M 5.7 168.0 GB 96% Q6_K 6.56 193.3 GB 97% Q8_0 8.5 250.3 GB 100% FP16 16 470.7 GB 100%
§ 01 BENCHMARK SCORES
HumanEval 95.0
MMLU-PRO 72.0
MATH 96.0
IFEval 88.0
GPQA 66.0
Arena Elo 1367.0
GPQA Diamond 75.3
LiveCodeBench 52.4
AIME 71.7
MATH-500 98.0
HLE 10.6
AA Intelligence 25.0
AA Coding 22.1
AA Math 71.7
aa_ifbench 46.1
aa_terminal_bench 15.2
aa_tau2 33.3
aa_scicode 36.0
aa_lcr 31.2
§ 02 RUN COMMAND
Run Qwen3-235B-A22B locally with Ollama — needs 144.2 GB VRAM at Q4_K_M:
$ ollama run qwen3:235b
§ 03 COMPATIBLE GPUs
13 @ Q4_K_M Feedback 