LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Inkling Small
◆ OPEN
by Thinkingmachines
OPEN WEIGHTSLOCALAPILONG CONTEXTPAIDFREE TO TRY
RANKED #10 OF 29 VIDEO · OVERALL #138 OF 504 RANKED · VIBE SCORE 5.7 · 64 VOICES ·
COPY BADGE pastes a README snippet with the rank SVG.
Inkling Small is an open-weight multimodal mixture-of-experts model from Thinking Machines Lab, with 12B active parameters out of 276B total. It is positioned as the smaller, more efficient member of...
HAPPY10.0
9 mentions
SAD5.0
1 mentions
VIBE TRAJECTORY
26 weeks · 93 voices in this window
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 Inkling Small once and it's yours. No subscription, no rate limits, works offline.
276B total parameters (12B active MoE), a Q4 / NVFP4 quantized checkpoint requires about 160 GB to 180 GB memory
- BARE MINIMUM
Apple Mac Studio M2 Ultra (192 GB Unified Memory)
192 GB UNIFIED MEMORY
~$5,500 USEDInkling Small is too large for standard consumer GPUs; requires high-memory workstation hardware to fit ~160 GB Q4/NVFP4
- SWEET SPOT
Workstation with 4x NVIDIA RTX 4090 (24 GB) and 256 GB Syste
96 GB VRAM + 256 GB RAM
~$9,500 CUSTOM BUILDQ4 / NVFP4 with hybrid GPU offloading, ~18 tok/s, 32k context
- FULL POWER
Dual NVIDIA H200 Workstation (141 GB SXM5 each)
282 GB HBM3E VRAM
~$70,000 ENTERPRISEFull precision BF16 or NVFP4 with full extended context up to 128k+, ~85 tok/s
LEARN HOW TO SET IT UP
WILL MY BOX RUN INKLING SMALL?
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 Inkling Small takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMApple Mac Studio M2 Ultra (192 GB Unified Memory) | SWEET SPOTWorkstation with 4x NVIDIA RTX 4090 (24 GB) and 256 GB Syste | FULL POWERDual NVIDIA H200 Workstation (141 GB SXM5 each) |
|---|---|---|---|
| Make a video clipfive seconds of footage | —★★★★★ | —★★★★★ | —★★★★★ |
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
People generally like Inkling Small, with some gripes. Biggest praise: speed.
56 POSTS · 8 COMMENTS · REDDIT · GITHUB · X · BLOG · BLUESKY
PEOPLE LOVE
- People feel it is worth the money (3 mentions)
- Answers come back fast (3 mentions)
PEOPLE COMPLAIN ABOUT
- People find it expensive (2 mentions)
TOPIC BY TOPIC
- PriceMOSTLY POSITIVE
3 thumbs up · 2 thumbs down
- SpeedPEOPLE LOVE THIS
3 thumbs up · 1 thumbs down
LOOKING FOR AN ALTERNATIVE?
Other models the crowd has fully reviewed, starting with video models like this one.
Grok Imagine
Runway Gen 4 Turbo
Seedance 2.5
LTX 2.3> SOURCES (12)
- REDDIT · 2026-07-30 · 156 UPVOTESInkling-Small is out now!
Inkling-Small is out now!
- X · 2026-08-19 · 123 UPVOTESIf you have 2 DGX Sparks and work a lot with vision models, and want something with: - audio - video - image - agentic -
If you have 2 DGX Sparks and work a lot with vision models, and want something with: - audio - video - image - agentic -
- REDDIT · 2026-07-30 · 121 UPVOTESInkling-Small-276B-12B, effort "max" VS Qwen3.6-27B
Inkling-Small-276B-12B, effort "max" VS Qwen3.6-27B
- REDDIT · 2026-08-05 · 67 UPVOTESInkling-Small 276B-A12B at ~2.9 tok/s on <10gb memory
Inkling-Small 276B-A12B at ~2.9 tok/s on <10gb memory
- REDDIT · 2026-07-31 · 51 UPVOTESWe've gotten some great medium sized models lately (DSV4 Flash 0731, Inkling Small, Laguna S 2.1, Step 3.7 Flash) but does anybody else…
We've gotten some great medium sized models lately (DSV4 Flash 0731, Inkling Small, Laguna S 2.1, Step 3.7 Flash) but does anybody else want to see some new 70-80b contenders?
- REDDIT · 2026-07-30 · 39 UPVOTESThinking Machine's smaller "Inkling Small" Artificial Analysis results
Thinking Machine's smaller "Inkling Small" Artificial Analysis results
- BLOG · 2026-08-14 · 32 UPVOTESGLM-5.3: How Chinese labs keep stride with the frontier
Housekeeping: I’m traveling so cannot make a voiceover for this post. EDIT — I added a bullet point 5 on the Chinese data industry after sending the email out. Today, Z.ai announced their GLM-5.3 model, currently only available in the coding plan, coming soon…
- BLOG · 2026-07-22 · 32 UPVOTESOpen models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next
Exciting news! My book trying to share post-training knowledge with the world is done and shipping soon. Order on Manning or Amazon . Thanks for the support. It’s currently the #1 AI book on Amazon :). Nathan and Florian sit down to discuss everything…
- BLOG · 2026-07-31 · 32 UPVOTES[AINews] GPT 5.6 price cut by 20%-80%: Cost of GPT 5.4 Intelligence dropped 13x in 4 months due to GPT 5.6 recursive self-optimization
One of our big “hero charts” a year ago (eventually adopted by Demis ) made the stunning observation that, holding LMSys Elo constant, GPT4 level intelligence fell by 1000x over 18 months : Back then, it was unclear if these were “noob gains” - going from…
- BLOG · 2026-08-17 · 32 UPVOTESTeaching Everyone to Fish for Tokens
Housekeeping: No voiceover for this post as I’m traveling. The oldest comparison people try to make is how what’s happening with open models compares to foundational open-source software projects like the Linux operating system. There are fairly clean…
- BLOG · 2026-08-11 · 32 UPVOTESNVIDIA and Local AI Community Fuel Open Source Models and Intelligent Agents
The open source ecosystem is making it easier for AI enthusiasts and developers to build, customize and run increasingly capable agents locally. Throughout August, NVIDIA is celebrating the partners and open source communities moving local AI forward, along…
- REDDIT · 2026-07-31 · 30 UPVOTESNow Suddenly too many choices for DGX Spark with Qwen 3.5 122B . What would be the next upgrade?
Now Suddenly too many choices for DGX Spark with Qwen 3.5 122B . What would be the next upgrade?