Hugging FaceTakes in pictures as well as text, and supports 201 languages and dialects.
- Rating, estimated
- 3.1
- Downloads
- 1.9M
- Parameters
- 4B
- Context
- 256K
- Download
- 2.2 to 9.1 GB
- Commercial use
- Yes
Runs great on M4 · 16 GB
Outmake picks the Q6_K size: a 4.2 GB download that uses 7.3 GB of memory, and writes about 14 words (19 tokens) a second.
About
This release went through post-training as well as pre-training. In Outmake it can call tools and think before it answers.
How it compares
Of the 7 chat models Outmake recommends for an 8 GB Mac, it was the 2nd most downloaded on Hugging Face last month.
Outmake estimates it is the 4th fastest of those 7 models, about twice as fast as K2-Horizon 3.7B.
Good to know
The maker says training on pictures was nearly as efficient as training on text alone.
Part of its training was scaled across environments with a million agents, on tasks that grew harder.
Sources: Qwen/Qwen3.5-4B model card, Outmake's speed estimates, LM Arena and Hugging Face. Every line comes from one of these.
Good at
Licence
Apache 2.0. Commercial use: yes. The maker's page has the full terms.
Qwen3.5 4B runs well from a Mac with an M1 and 8 GB of memory, at Q3_K_M, writing about 11 words (14 tokens) a second. 59 of 59 Mac configurations run it well.
Compare with
The nearest models in size, judged for the same Mac.
- K2-Horizon 3.7BFor tasks that need tools called along the way, with reasoning worked through before each answer.Runs great · 4.4 GB · 7.4 words/sGet
- Gemma 4 E2BChat with photos and audioRuns great · 5.5 GB · 14 words/sGet
- Qwen3.5 2BReads text and picturesRuns great · 2.2 GB · 31 words/sGet
- Gemma 4 E4BUnderstands photos and speechRuns great · 7.6 GB · 8.9 words/sGet
Sizes
| Size | Download | Memory at 8K | Memory at 32K | Open |
|---|---|---|---|---|
| Q3_K_MSmaller | 3.0 GB | 3.7 GB | 4.1 GB | Open |
| Q4_K_MBalanced | 3.4 GB | 4.1 GB | 4.6 GB | Open |
| Q6_KBest | 4.2 GB | 4.9 GB | 5.4 GB | Open |
All 22 sizes
| Size | Download | Memory at 8K | Memory at 32K | Open |
|---|---|---|---|---|
| UD-IQ2_XXS | 2.2 GB | 2.9 GBestimated | 3.4 GB | Open |
| UD-IQ2_M | 2.4 GB | 3.1 GBestimated | 3.6 GB | Open |
| UD-Q2_K_XL | 2.6 GB | 3.3 GBestimated | 3.8 GB | Open |
| UD-IQ3_XXS | 2.6 GB | 3.3 GBestimated | 3.8 GB | Open |
| Q3_K_S | 2.8 GB | 3.5 GBestimated | 3.9 GB | Open |
| UD-Q3_K_XL | 3.1 GB | 3.8 GBestimated | 4.3 GB | Open |
| IQ4_XS | 3.1 GB | 3.9 GBestimated | 4.3 GB | Open |
| IQ4_NL | 3.3 GB | 4.0 GBestimated | 4.4 GB | Open |
| Q4_0 | 3.3 GB | 4.0 GBestimated | 4.4 GB | Open |
| Q4_K_S | 3.3 GB | 4.0 GBestimated | 4.4 GB | Open |
| Q4_1 | 3.5 GB | 4.2 GBestimated | 4.6 GB | Open |
| UD-Q4_K_XL | 3.6 GB | 4.3 GBestimated | 4.7 GB | Open |
| Q5_K_S | 3.7 GB | 4.4 GBestimated | 4.9 GB | Open |
| Q5_K_M | 3.8 GB | 4.5 GBestimated | 5.0 GB | Open |
| UD-Q5_K_XL | 3.9 GB | 4.6 GBestimated | 5.1 GB | Open |
| UD-Q6_K_XL | 4.8 GB | 5.5 GBestimated | 6.0 GB | Open |
| Q8_0 | 5.2 GB | 5.9 GBestimated | 6.3 GB | Open |
| UD-Q8_K_XL | 6.6 GB | 7.3 GBestimated | 7.8 GB | Open |
| BF16 | 9.1 GB | 9.8 GBestimated | 10 GB | Open |
Memory is what llama.cpp's own calculator reported for each size when the catalog was built, with the app's settings. Smaller, Balanced and Best are the three sizes the store names.
Run It Anywhere
Qwen3.5 4B is an open model, so any app that runs GGUF files can run it. These take the Q4_K_M size from unsloth straight from Hugging Face.
- LM StudioMac, Windows, Linux
Open - JanMac, Windows, Linux
Open - MstyMac, Windows, Linux
Open - OllamaMac, Windows, Linux
ollama run hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M - llama.cppMac, Windows, Linux
llama-server -hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M - Docker Model RunnerMac, Windows, Linux
docker model run hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M
On a PC Graphics Card
| Video memory | Largest size that fits | Download | Memory at 8K |
|---|---|---|---|
| 6 GB | Q8_0 | 5.2 GB | 5.9 GB |
| 8 GB | UD-Q8_K_XL | 6.6 GB | 7.3 GB |
| 10 GB | BF16 | 9.1 GB | 9.8 GB |
| 11 GB | BF16 | 9.1 GB | 9.8 GB |
| 12 GB | BF16 | 9.1 GB | 9.8 GB |
| 16 GB | BF16 | 9.1 GB | 9.8 GB |
| 20 GB | BF16 | 9.1 GB | 9.8 GB |
| 24 GB | BF16 | 9.1 GB | 9.8 GB |
| 32 GB | BF16 | 9.1 GB | 9.8 GB |
| 48 GB | BF16 | 9.1 GB | 9.8 GB |
The largest size, of those that keep three bits a weight or more, whose memory at an 8K chat is under 92% of the card's; the rest is the desktop's. Estimated from the files, the same way as for a Mac; speed on a PC is not estimated yet. A size that doesn't fit the card still runs, slower, with part of it in ordinary memory.
On every Mac
| Mac | Size | Fit | Memory | Writes | Reads |
|---|---|---|---|---|---|
| M6 · 16 GB | Q6_K | Runs great | 7.3 GB | 21 | 1572 |
| M6 · 24 GB | Q6_K | Runs great | 7.3 GB | 21 | 1572 |
| M6 · 32 GB | Q6_K | Runs great | 7.3 GB | 21 | 1572 |
| M5 · 16 GB | Q6_K | Runs great | 7.3 GB | 19 | 580 |
| M5 · 24 GB | Q6_K | Runs great | 7.3 GB | 19 | 580 |
| M5 · 32 GB | Q6_K | Runs great | 7.3 GB | 19 | 580 |
| M5 Pro · 24 GB | Q6_K | Runs great | 7.3 GB | 37 | 1161 |
| M5 Pro · 48 GB | Q6_K | Runs great | 7.3 GB | 37 | 1161 |
| M5 Pro · 64 GB | Q6_K | Runs great | 7.3 GB | 37 | 1451 |
| M5 Max · 36 GB | Q6_K | Runs great | 7.3 GB | 54 | 2322 |
| M5 Max · 48 GB | Q6_K | Runs great | 7.3 GB | 69 | 2902 |
| M5 Max · 64 GB | Q6_K | Runs great | 7.3 GB | 69 | 2902 |
| M5 Max · 128 GB | Q6_K | Runs great | 7.3 GB | 69 | 2902 |
| M5 Ultra · 96 GB | Q6_K | Runs great | 7.3 GB | 122 | 4644 |
| M5 Ultra · 256 GB | Q6_K | Runs great | 7.3 GB | 122 | 5804 |
| M5 Ultra · 512 GB | Q6_K | Runs great | 7.3 GB | 122 | 5804 |
| M4 · 16 GB | Q6_K | Runs great | 7.3 GB | 14 | 195 |
| M4 · 24 GB | Q6_K | Runs great | 7.3 GB | 14 | 195 |
| M4 · 32 GB | Q6_K | Runs great | 7.3 GB | 14 | 195 |
| M4 Pro · 24 GB | Q6_K | Runs great | 7.3 GB | 28 | 390 |
| M4 Pro · 48 GB | Q6_K | Runs great | 7.3 GB | 28 | 390 |
| M4 Pro · 64 GB | Q6_K | Runs great | 7.3 GB | 28 | 390 |
| M4 Max · 36 GB | Q6_K | Runs great | 7.3 GB | 38 | 780 |
| M4 Max · 48 GB | Q6_K | Runs great | 7.3 GB | 46 | 975 |
| M4 Max · 64 GB | Q6_K | Runs great | 7.3 GB | 46 | 975 |
| M4 Max · 96 GB | Q6_K | Runs great | 7.3 GB | 38 | 780 |
| M4 Max · 128 GB | Q6_K | Runs great | 7.3 GB | 46 | 975 |
| M3 · 8 GB | Q3_K_M | Runs great | 3.7 GB | 16 | 161 |
| M3 · 16 GB | Q6_K | Runs great | 7.3 GB | 12 | 161 |
| M3 · 24 GB | Q6_K | Runs great | 7.3 GB | 12 | 161 |
| M3 Pro · 18 GB | Q6_K | Runs great | 7.3 GB | 17 | 283 |
| M3 Pro · 36 GB | Q6_K | Runs great | 7.3 GB | 17 | 283 |
| M3 Max · 36 GB | Q6_K | Runs great | 7.3 GB | 31 | 605 |
| M3 Max · 48 GB | Q6_K | Runs great | 7.3 GB | 39 | 807 |
| M3 Max · 64 GB | Q6_K | Runs great | 7.3 GB | 39 | 807 |
| M3 Max · 96 GB | Q6_K | Runs great | 7.3 GB | 31 | 605 |
| M3 Max · 128 GB | Q6_K | Runs great | 7.3 GB | 39 | 807 |
| M3 Ultra · 96 GB | Q6_K | Runs great | 7.3 GB | 62 | 1211 |
| M3 Ultra · 256 GB | Q6_K | Runs great | 7.3 GB | 62 | 1211 |
| M3 Ultra · 512 GB | Q6_K | Runs great | 7.3 GB | 62 | 1614 |
| M2 · 8 GB | Q3_K_M | Runs great | 3.7 GB | 16 | 158 |
| M2 · 16 GB | Q6_K | Runs great | 7.3 GB | 12 | 158 |
| M2 · 24 GB | Q6_K | Runs great | 7.3 GB | 12 | 158 |
| M2 Pro · 16 GB | Q6_K | Runs great | 7.3 GB | 22 | 317 |
| M2 Pro · 32 GB | Q6_K | Runs great | 7.3 GB | 22 | 317 |
| M2 Max · 32 GB | Q6_K | Runs great | 7.3 GB | 39 | 594 |
| M2 Max · 64 GB | Q6_K | Runs great | 7.3 GB | 39 | 594 |
| M2 Max · 96 GB | Q6_K | Runs great | 7.3 GB | 39 | 753 |
| M2 Ultra · 64 GB | Q6_K | Runs great | 7.3 GB | 62 | 1189 |
| M2 Ultra · 128 GB | Q6_K | Runs great | 7.3 GB | 62 | 1189 |
| M2 Ultra · 192 GB | Q6_K | Runs great | 7.3 GB | 62 | 1189 |
| M1 · 8 GB | Q3_K_M | Runs great | 3.7 GB | 11 | 128 |
| M1 · 16 GB | Q6_K | Runs great | 7.3 GB | 8 | 128 |
| M1 Pro · 16 GB | Q6_K | Runs great | 7.3 GB | 20 | 257 |
| M1 Pro · 32 GB | Q6_K | Runs great | 7.3 GB | 20 | 257 |
| M1 Max · 32 GB | Q6_K | Runs great | 7.3 GB | 33 | 441 |
| M1 Max · 64 GB | Q6_K | Runs great | 7.3 GB | 33 | 441 |
| M1 Ultra · 64 GB | Q6_K | Runs great | 7.3 GB | 49 | 881 |
| M1 Ultra · 128 GB | Q6_K | Runs great | 7.3 GB | 49 | 881 |
Writes and reads are words a second (a token is about three quarters of a word), 4K tokens into a chat, estimated unless marked measured. The size is the one the store offers that Mac. Constants fitted October 11, 2026; how, and how wrong.
Badge for the model card
Shows on Qwen3.5 4B's Hugging Face card which Macs run it well, linked to this page. It updates every morning.
Sizes and files are read from unsloth/Qwen3.5-4B-GGUF on Hugging Face and checked every morning. LM Arena has not rated this model, so the rating is Outmake's estimate from its size and release date; how the estimate is made. Catalog of November 4, 2026.