# Strata

31 Hacker News commenters recommended Strata and 15 criticized it, November 2025 to November 2026. It was recommended in 5 of 13 months. It meets the Hacker Approved badge rule.

**Verdict:** Runs large local LLMs fast on consumer hardware. Caveats: not open source, and quantized models may make more mistakes.

## Why people recommend it

- Runs big models on modest hardware (8 comments): "It will run reasonably fast if you have 64+ gb ram and use strata or a strata fork" ([source](https://news.ycombinator.com/item?id=50161268)) "You can get something relatively capable and usable with your configuration" ([source](https://news.ycombinator.com/item?id=50148692)) "It is fully possible to run the 180B Owen model on strix halo using strata" ([source](https://news.ycombinator.com/item?id=50019856))
- Faster token speeds than llama.cpp (6 comments): "the results are not even close. You can basically ignore everything in that table and use Strata instead" ([source](https://news.ycombinator.com/item?id=50262022)) "I've switched to IQ3_XXS and am running at 60 t/s on Strata vs the 21 t/s I was getting in llama.cpp." ([source](https://news.ycombinator.com/item?id=49955250)) "Strata has really shown how much performance is left on the table" ([source](https://news.ycombinator.com/item?id=50136321))
- Stable with no crashes (2 comments): "its been running on my system for days without any crash" ([source](https://news.ycombinator.com/item?id=50150209)) "without inference crashing before long" ([source](https://news.ycombinator.com/item?id=50149101))
- Opens huge log files quickly (2 comments): "it takes 8 seconds to open a 50GB file and once you do scrolling is buttery smooth with 60fps" ([source](https://news.ycombinator.com/item?id=46302649)) "can open a 50GB log file in 8 seconds and not eat your RAM" ([source](https://news.ycombinator.com/item?id=46300743))

## What people warn about

- Not open source (2 comments): "though I'd find it much more compelling if it were OSS" ([source](https://news.ycombinator.com/item?id=49913459)) "I'd find it much more compelling if it were OSS." ([source](https://news.ycombinator.com/item?id=49915296))
- Quantization hurts output quality (3 comments): "they quantize, worse they quantize KV caches, they approximate expert routing etc." ([source](https://news.ycombinator.com/item?id=50313634)) "perhaps such metric is the most challenging one for a brutally quantized and otherwise lobotomized model" ([source](https://news.ycombinator.com/item?id=49962456)) "same model running on Strata made more mistakes and hallucinated more when compare to llama.cpp" ([source](https://news.ycombinator.com/item?id=49964809))
- Dangerous for outdoor safety decisions (5 comments): "I saw no evidence of trials to test to see how the technology performs against real conditions." ([source](https://news.ycombinator.com/item?id=48812922)) "Llms never took a difficult on field decision, never had to turn gut feeling into calls." ([source](https://news.ycombinator.com/item?id=48814337)) "This is going to kill someone... you can't just make these calls for people." ([source](https://news.ycombinator.com/item?id=48812789))

## Alternatives hackers recommend

- [vLLM](https://hackerapproved.com/p/vllm.md): 112 recommend
- [DwarfStar 4](https://hackerapproved.com/p/dwarfstar.md): 26 recommend
- [SGLang](https://hackerapproved.com/p/sglang.md): 19 recommend
- [ninfer](https://hackerapproved.com/p/ninfer.md): 16 recommend
- [exo](https://hackerapproved.com/p/exo.md): 15 recommend

## Cite as

Hacker Approved, "Strata", https://hackerapproved.com/p/strata/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).