DwarfStar 4
Hacker ApprovedRanked 3 of 27 inference engines by recommenders
Lets people run large DeepSeek-class models locally on consumer hardware, supporting few models well. Critics say speeds are often too slow for comfort.
Lets people run large DeepSeek-class models locally on consumer hardware, supporting few models well. Critics say speeds are often too slow for comfort.
28 people gave an opinion
- recommend it
- 26
- criticize it
- 2
- approval share
- 93%
Recommended in 6 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 26 people recommend itNeeds 10 or more, met.
- 13 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 6 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Runs big models on consumer hardware
“Projects like DwarfStar really lower the hardware bar a lot so Deepseek 4.1 flash and other mixture of expert models can run on consumer hardware.”
Comment from 8 Oct 2026 in “Why isn't the industry freaking out about DeepSeek 4.1 Flash?” “the amazing work being done by antirez on the DwarfStar 4 to run agents on a quasi-frontier class model on consumer hardware”
Comment from 19 May 2026 in “What can a local model do for you in early May 2026?” “DwarfStar is the only thing I've run that doesn't try and make my Mac Studio 128GB take off.”
Comment from 30 Jun 2026 in “Qwen 3.6 27B is the sweet spot for local development” - Comment from 17 May 2026 in “Apple Silicon costs more than OpenRouter”
- Comment from 29 Jun 2026 in “Qwen 3.6 27B is the sweet spot for local development”
Usable for real work
“this project is much more than PoC and actually both usable and useful for people doing local inference”
Comment from 11 Sep 2026 in “Astra for Coding: Why Are We Doing This Again?” “Not only for toying, but for getting work done.”
Comment from 14 Aug 2026 in “GLM-5.3: Frontier Coding with Emergent Cyber Capabilities” “DS4 is designed to do real-work.”
Comment from 29 Jul 2026 in “Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac” - Comment from 9 Sep 2026 in “DeepSeek launching v4.1 flash cheaper and more capable than v4 pro”
Fast, reasonable speeds
“you can run DeepSeek V4 flash at good speed via DwarfStar”
Comment from 3 Jul 2026 in “Jamesob's guide to running SOTA LLMs locally” “the fastest inference engine out there”
Comment from 17 Oct 2026 in “The Commodore 64 hardware simulator: break chips and pins, see the results” “at reasonable speeds and quality”
Comment from 22 Jul 2026 in “Laguna S 2.1” Supports few models well
“it only supports a small set of carefully chosen models, but it supports them really well”
Comment from 2 Oct 2026 in “From the creator of Redis; run LLM locally with ds4” “Antirez' DwarfStar supports it really well.”
Comment from 23 Jul 2026 in “New Framework Desktop Option with AMD Ryzen AI Max+ Pro 495 and 192GB Memory”
What people warn about
Each mark is one comment
Speeds too slow for comfort
“it's really hard for me to personally go back to speeds like that”
Comment from 10 Aug 2026 in “Meta Muse Glimmer – open weights 30B local coding model” “it is usable but it is far from the 35 t/s 400 t/s prefill”
Comment from 23 Jun 2026 in “Unsloth GLM-5.2 – How to Run Locally” “it's not reasonable speed, at all”
Comment from 23 Jul 2026 in “Laguna S 2.1”
Other inference engines people recommend
The four most recommended other inference engines, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
See all 27 inference engines, rankedWhat people said
34 of the 36 comments that mention it, newest first. Each links to the original comment.
“pretty good - decent TPS and coding ability”
“the fastest inference engine out there”
“Projects like DwarfStar really lower the hardware bar a lot so Deepseek 4.1 flash and other mixture of expert models can run on consumer hardware.”
“I use the q4 quant daily and it works really well.”
“it only supports a small set of carefully chosen models, but it supports them really well”
“they are terrific projects with extremely strong architecture”
“this project is much more than PoC and actually both usable and useful for people doing local inference”
“I didn't see anything that would make me want to stick with it vs llama.cpp”
“it's been very solid for me”
“Not only for toying, but for getting work done.”
“it's really hard for me to personally go back to speeds like that”
“This with DwarfStar delivers usable local AI (I hope!)”
“It should work on your 64 GB after you enable SSD mode in DwarfStar”
“DS4 is designed to do real-work.”
“Antirez' DwarfStar supports it really well.”
“it's not reasonable speed, at all”
“at reasonable speeds and quality”
“feels like local llm "just works" for reasonably decent coding”
“you can run DeepSeek V4 flash at good speed via DwarfStar”
“DeepSeek-4 Flash Q2 - last 6 layers Q4 quant with DwarfStar which just about fits in 128Gb is definitely superior IMO”
“DwarfStar is the only thing I've run that doesn't try and make my Mac Studio 128GB take off.”
“Try using DwarfStar 4 and use the --power flag”
“DeepSeek-V4-Flash will fit, served by DwarfStar”
“Thank you for your work on DwarfStar! It is truly helping democratize access to frontier tech.”
“it is usable but it is far from the 35 t/s 400 t/s prefill”
“somehow seems to not suffer as much as I'd thought”
“that was enough of a taste for me to jump into 2 sparks”
“the amazing work being done by antirez on the DwarfStar 4 to run agents on a quasi-frontier class model on consumer hardware”
“Tho says Metal / CUDA, so doesn't seem friendly to Linux AMD system”
“DwarfStar 4 making it easy to run on a 96GB MacBook”
“with DwarfStar steering features I was able to completely remove refusal from DS4”
“Been a lot of fun to play around with it”
“See DwarfStar 4 for current sweet spots”
“The long context reasoning is something I haven't even seen in frontier models”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.