LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
DeepSeek V3.1
◆ OPEN
by DeepSeek
OPEN WEIGHTSLOCALAPIPAIDCLOSEDREASONING
RANKED #103 OF 228 ASSISTANTS · OVERALL #175 OF 6,618 · VIBE SCORE 5.9 · 234 VOICES
DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context...
HAPPY8.6
38 mentions
SAD6.4
16 mentions
VIBE TRAJECTORY
26 weeks · 56 voices
HOLDING UP
Recent vibe is steady. No clear fade in the crowd.
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 DeepSeek V3.1 once and it's yours. No subscription, no rate limits, works offline.
671B parameters (37B active MoE), a dynamic 2-bit quant is ~245 GB while standard Q4 requires ~400 GB
- BARE MINIMUM
Mac Studio M2 Ultra (192 GB Unified Memory)
192 GB RAM
~$5000 NEWToo large for consumer GPUs; runs dynamic 1.5-bit to 2-bit quants with CPU memory swapping, ~3 to 5 tok/s, short context
- SWEET SPOT
Mac Studio M3 Ultra or dual Mac Studio cluster with 256+ GB
256 GB RAM
~$7500 USEDDynamic 2.5-bit to 3-bit GGUF, ~8 to 12 tok/s, 16k context
- FULL POWER
Custom 8x Nvidia RTX 6000 Ada workstation (384 GB VRAM) or 8
384 GB VRAM
~$55000 NEWFP8 native precision, 40+ tok/s, full 128k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN DEEPSEEK V3.1?
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 DeepSeek V3.1 takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMMac Studio M2 Ultra (192 GB Unified Memory) | SWEET SPOTMac Studio M3 Ultra or dual Mac Studio cluster with 256+ GB | FULL POWERCustom 8x Nvidia RTX 6000 Ada workstation (384 GB VRAM) or 8 |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~2.1 MIN★★★★★ | ~50 S★★★★★ | —★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~6.3 MIN★★★★★ | ~2.5 MIN★★★★★ | —★★★★★ |
| Build a websitea small landing page, markup and styles together | ~33 MIN★★★★★ | ~13 MIN★★★★★ | —★★★★★ |
| Build a backendan API with routes, storage and tests | ~1.7 HR★★★★★ | ~42 MIN★★★★★ | —★★★★★ |
| Build a gamea playable browser game in one file | ~63 MIN★★★★★ | ~25 MIN★★★★★ | —★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. A DASH MEANS NO SPEED WAS RESEARCHED FOR THAT RIG. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
The internet is split on DeepSeek V3.1. Biggest praise: price. Biggest gripe: getting facts right.
153 POSTS · 81 COMMENTS · HACKER NEWS · X · DEV.TO · REDDIT · LEMMY · GITHUB · BLOG · DEV FORUMS
PEOPLE LOVE
- People feel it is worth the money (3 mentions)
PEOPLE COMPLAIN ABOUT
- Sometimes makes things up (3 mentions)
TOPIC BY TOPIC
- PricePEOPLE LOVE THIS
3 thumbs up · 1 thumbs down
- Getting facts rightA SORE SPOT
0 thumbs up · 3 thumbs down
LOOKING FOR AN ALTERNATIVE?
Checked picks first: tags say why you would switch, stars say how fully each one stands in for DeepSeek V3.1.
GLM 5.3> SOURCES (12)
- REDDIT · 2026-02-17 · 587 UPVOTESCar Wash Test on 53 leading models: “I want to wash my car. The car wash is 50 meters away. Should I walk or drive?”
Car Wash Test on 53 leading models: “I want to wash my car. The car wash is 50 meters away. Should I walk or drive?”
- REDDIT · 2025-08-21 · 513 UPVOTESDeepSeek-V3.1 has officially launched
DeepSeek-V3.1 has officially launched
- REDDIT · 2026-03-02 · 505 UPVOTESDeepseek V4 - All Leaks and Infos for the Release Day - Not Verified!
Deepseek V4 - All Leaks and Infos for the Release Day - Not Verified!
- REDDIT · 2026-01-25 · 385 UPVOTESOllama Models Ranked by VRAM Requirements
Ollama Models Ranked by VRAM Requirements
- REDDIT · 2025-08-19 · 326 UPVOTESDeepSeek v3.1 already does better than ChatGPT-5. Change my mind.
DeepSeek v3.1 already does better than ChatGPT-5. Change my mind.
- REDDIT · 2025-08-19 · 305 UPVOTESDeepSeek v3.1 just went live on HuggingFace
DeepSeek v3.1 just went live on HuggingFace
- REDDIT · 2025-07-11 · 288 UPVOTESKimi K2: New SoTA non-reasoning model 1T parameters open-source and outperforms DeepSeek-v3.1 and GPT-4.1 by a large margin
Kimi K2: New SoTA non-reasoning model 1T parameters open-source and outperforms DeepSeek-v3.1 and GPT-4.1 by a large margin
- REDDIT · 2025-08-19 · 221 UPVOTESDeepSeek v3.1 just went live on HuggingFace
DeepSeek v3.1 just went live on HuggingFace
- REDDIT · 2025-11-22 · 217 UPVOTESYour local Ollama agents can be just as good as closed-source models - I open-sourced Stanford's ACE framework that makes agents learn from…
Your local Ollama agents can be just as good as closed-source models - I open-sourced Stanford's ACE framework that makes agents learn from mistakes
- LEMMY · 2025-10-24 · 204 UPVOTESAI sycophancy (excessively agreeing with user) is pervasive and harmful for people who seek advice from AIs
AI sycophancy (excessively agreeing with user) is pervasive and harmful for people who seek advice from AIs
- REDDIT · 2025-09-24 · 185 UPVOTESYou can now run DeepSeek-V3.1-Terminus locally!
You can now run DeepSeek-V3.1-Terminus locally!
- REDDIT · 2026-01-22 · 167 UPVOTESV4 is coming soon
V4 is coming soon