LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Qwen3 235B A22B
◆ OPEN
by Alibaba
OPEN WEIGHTSLOCALAGENTSWRITING HELPCODING HELP
RANKED #36 OF 267 ASSISTANTS · OVERALL #55 OF 6,566 · VIBE SCORE 6.7 · 1,320 VOICES
A powerful open AI model designed to handle complex tasks and provide accurate answers.
HAPPY9.2
178 mentions
SAD5.8
100 mentions
VIBE TRAJECTORY
26 weeks · 1,025 voices
QUIETLY DEGRADING
Recent vibe and chatter are both sliding.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Qwen3 235B A22B once and it's yours. No subscription, no rate limits, works offline.
235B parameters (MoE with 22B active), a Q4 file is about 143 GB, while a Q2 file is about 88 GB
- BARE MINIMUM
Apple Mac Studio M2 Ultra (used)
128 GB UNIFIED MEMORY
~$2500 USEDQ2 (UD-Q2_K_XL) or Q3 quant, ~5 to 8 tok/s, 32k context
- SWEET SPOT
Custom PC with 3x RTX 3090 GPUs (used) and Threadripper CPU
72 GB VRAM AND 128 GB RAM
~$3600 BUILTQ3_K_XL or Q4_K_M quant, ~15 tok/s, 32k context
- FULL POWER
Apple Mac Studio M3 Ultra
256 GB UNIFIED MEMORY
~$6000 NEWQ8 or FP8 precision, ~10 to 12 tok/s, up to 256k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN QWEN3 235B A22B?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Qwen3 235B A22B takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMApple Mac Studio M2 Ultra (used) | SWEET SPOTCustom PC with 3x RTX 3090 GPUs (used) and Threadripper CPU | FULL POWERApple Mac Studio M3 Ultra |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~77 S★★★★★ | ~33 S★★★★★ | ~45 S★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~3.8 MIN★★★★★ | ~1.7 MIN★★★★★ | ~2.3 MIN★★★★★ |
| Build a websitea small landing page, markup and styles together | ~21 MIN★★★★★ | ~8.9 MIN★★★★★ | ~12 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~64 MIN★★★★★ | ~28 MIN★★★★★ | ~38 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~38 MIN★★★★★ | ~17 MIN★★★★★ | ~23 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
Users love that Qwen3 235B A22B consistently delivers correct answers and excels at multi-agent tasks. However, many people find it expensive to run.
251 POSTS · 1069 COMMENTS · REDDIT · HACKER NEWS · BLUESKY · STACK OVERFLOW · OTHER FORUMS · LEMMY · DEV FORUMS · X · BLOG · GITHUB
PEOPLE LOVE
- Usually gives correct answers (3 mentions)
PEOPLE COMPLAIN ABOUT
- People find it expensive (2 mentions)
TOPIC BY TOPIC
- PriceMIXED FEELINGS
2 thumbs up · 2 thumbs down
- Getting facts rightPEOPLE LOVE THIS
3 thumbs up · 0 thumbs down
IN THEIR OWN WORDS
- FROM REDDIT >
“Built a 6-agent psychology pipeline on self-hosted Qwen3-235B — the jump from 30B was actually worth it for this task.”
LOOKING FOR AN ALTERNATIVE?
Checked picks first: tags say why you would switch, stars say how fully each one stands in for Qwen3 235B A22B.
DeepSeek V4 Pro
Grok 4.6> SOURCES (12)
- REDDIT · 2026-07-19 · 4099 UPVOTESQwen3.8
Qwen3.8
- HACKER NEWS · 2026-08-14 · 1438 UPVOTESQwen 3.8 27B
Qwen 3.8 27B
- HACKER NEWS · 2026-06-29 · 1192 UPVOTESQwen 3.6 27B is the sweet spot for local development
Qwen 3.6 27B is the sweet spot for local development
- HACKER NEWS · 2026-08-03 · 1124 UPVOTESQwen3.8-Max: A New Bar for Coding and Cowork
Qwen3.8-Max: A New Bar for Coding and Cowork
- HACKER NEWS · 2026-07-19 · 961 UPVOTESQwen 3.8
Qwen 3.8
- HACKER NEWS · 2026-08-16 · 802 UPVOTESQwen 3.8 27B is excellent, but it defaults to overthinking things
Qwen 3.8 27B is excellent, but it defaults to overthinking things
- HACKER NEWS · 2026-01-22 · 744 UPVOTESQwen3-TTS family is now open sourced: Voice design, clone, and generation
Qwen3-TTS family is now open sourced: Voice design, clone, and generation
- HACKER NEWS · 2026-09-03 · 689 UPVOTESQwen 3.8 27B available on Cerebras at 1500 tokens/s
Qwen 3.8 27B available on Cerebras at 1500 tokens/s
- HACKER NEWS · 2025-09-22 · 571 UPVOTESQwen3-Omni: Native Omni AI model for text, image and video
Qwen3-Omni: Native Omni AI model for text, image and video
- HACKER NEWS · 2025-09-12 · 569 UPVOTESQwen3-Next
Qwen3-Next
- HACKER NEWS · 2026-08-06 · 546 UPVOTESQwen3.8 Max now ranked as the best overall model by agentic index
Qwen3.8 Max now ranked as the best overall model by agentic index
- HACKER NEWS · 2026-02-28 · 461 UPVOTESQwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers
Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers