model family

Qwen 3

Published by Qwen. 9 members measured on real machines, across 28 builds. Every number below was signed by the machine that produced it — none of it is from a model card.

what it takes, what it returns
text → text

Declared by the network’s catalog, not measured — a signature says what this family would be asked. Whether it answered is the signed capability flags on each member.

28 of 54 builds measured · 26 still open
what each build proved
vl-4b · F16
vl-8b · F16
30b-a3b · Q4_K_M
1.7b
14b
4b
8b
4b · IQ3_XXS
4b · Q5_0
4b · Q5_K_M
4b · Q8_0
14b · Q8_0
4b · Q4_K_M
4b · BF16
4b · Q4_0
vl-8b~prithivmlmods-abliterated · Q2_K
4b · Q4_1 · unsloth
0.6b · Q4_K_M
0.6b · Q8_0
4b · IQ4_XS
0.6b · Q2_K
4b · IQ1_S
4b · Q4_1 · unsloth
0.6b · Q3_K_M
0.6b · Q5_K_M
1.7b · Q8_0
4b · Q2_K
0.6b
vl-4b · F16 · Tool calling — provedvl-4b · F16 · Tool loop finished — asked, failedvl-4b · F16 · Faithful to tool results — provedvl-4b · F16 · Chained tool calls — asked, failedvl-4b · F16 · Structured output — provedvl-4b · F16 · Sustained output — provedvl-4b · F16 · Extended thinking — asked, failedvl-4b · F16 · Vision — provedvl-4b · F16 · Counts in images — provedvl-4b · F16 · Multiple images at once — provedvl-4b · F16 · Spatial reasoning — provedvl-4b · F16 · Writes code — provedvl-4b · F16 · Reasons about code — asked, failedvl-4b · F16 · Synthesises sources — asked, failedvl-4b · F16 · Follows instructions — provedvl-8b · F16 · Tool calling — provedvl-8b · F16 · Tool loop finished — asked, failedvl-8b · F16 · Faithful to tool results — provedvl-8b · F16 · Chained tool calls — asked, failedvl-8b · F16 · Structured output — provedvl-8b · F16 · Sustained output — provedvl-8b · F16 · Extended thinking — asked, failedvl-8b · F16 · Vision — provedvl-8b · F16 · Counts in images — provedvl-8b · F16 · Multiple images at once — provedvl-8b · F16 · Spatial reasoning — provedvl-8b · F16 · Writes code — provedvl-8b · F16 · Reasons about code — asked, failedvl-8b · F16 · Synthesises sources — asked, failedvl-8b · F16 · Follows instructions — proved30b-a3b · Q4_K_M · Tool calling — asked, failed30b-a3b · Q4_K_M · Tool loop finished — never asked30b-a3b · Q4_K_M · Faithful to tool results — never asked30b-a3b · Q4_K_M · Chained tool calls — never asked30b-a3b · Q4_K_M · Structured output — proved30b-a3b · Q4_K_M · Sustained output — proved30b-a3b · Q4_K_M · Extended thinking — asked, failed30b-a3b · Q4_K_M · Vision — asked, failed30b-a3b · Q4_K_M · Counts in images — never asked30b-a3b · Q4_K_M · Multiple images at once — never asked30b-a3b · Q4_K_M · Spatial reasoning — never asked30b-a3b · Q4_K_M · Writes code — proved30b-a3b · Q4_K_M · Reasons about code — asked, failed30b-a3b · Q4_K_M · Synthesises sources — proved30b-a3b · Q4_K_M · Follows instructions — proved1.7b · Tool calling — proved1.7b · Tool loop finished — asked, failed1.7b · Faithful to tool results — proved1.7b · Chained tool calls — proved1.7b · Structured output — proved1.7b · Sustained output — proved1.7b · Extended thinking — proved1.7b · Vision — never asked1.7b · Counts in images — never asked1.7b · Multiple images at once — never asked1.7b · Spatial reasoning — never asked1.7b · Writes code — asked, failed1.7b · Reasons about code — proved1.7b · Synthesises sources — proved1.7b · Follows instructions — asked, failed14b · Tool calling — proved14b · Tool loop finished — asked, failed14b · Faithful to tool results — proved14b · Chained tool calls — proved14b · Structured output — proved14b · Sustained output — proved14b · Extended thinking — proved14b · Vision — never asked14b · Counts in images — never asked14b · Multiple images at once — never asked14b · Spatial reasoning — never asked14b · Writes code — asked, failed14b · Reasons about code — proved14b · Synthesises sources — proved14b · Follows instructions — asked, failed4b · Tool calling — proved4b · Tool loop finished — asked, failed4b · Faithful to tool results — proved4b · Chained tool calls — proved4b · Structured output — proved4b · Sustained output — proved4b · Extended thinking — proved4b · Vision — never asked4b · Counts in images — never asked4b · Multiple images at once — never asked4b · Spatial reasoning — never asked4b · Writes code — asked, failed4b · Reasons about code — proved4b · Synthesises sources — proved4b · Follows instructions — asked, failed8b · Tool calling — proved8b · Tool loop finished — asked, failed8b · Faithful to tool results — proved8b · Chained tool calls — proved8b · Structured output — proved8b · Sustained output — proved8b · Extended thinking — proved8b · Vision — never asked8b · Counts in images — never asked8b · Multiple images at once — never asked8b · Spatial reasoning — never asked8b · Writes code — asked, failed8b · Reasons about code — proved8b · Synthesises sources — proved8b · Follows instructions — asked, failed4b · IQ3_XXS · Tool calling — proved4b · IQ3_XXS · Tool loop finished — asked, failed4b · IQ3_XXS · Faithful to tool results — proved4b · IQ3_XXS · Chained tool calls — proved4b · IQ3_XXS · Structured output — proved4b · IQ3_XXS · Sustained output — proved4b · IQ3_XXS · Extended thinking — proved4b · IQ3_XXS · Vision — never asked4b · IQ3_XXS · Counts in images — never asked4b · IQ3_XXS · Multiple images at once — never asked4b · IQ3_XXS · Spatial reasoning — never asked4b · IQ3_XXS · Writes code — proved4b · IQ3_XXS · Reasons about code — proved4b · IQ3_XXS · Synthesises sources — proved4b · IQ3_XXS · Follows instructions — proved4b · Q5_0 · Tool calling — proved4b · Q5_0 · Tool loop finished — asked, failed4b · Q5_0 · Faithful to tool results — proved4b · Q5_0 · Chained tool calls — proved4b · Q5_0 · Structured output — proved4b · Q5_0 · Sustained output — proved4b · Q5_0 · Extended thinking — proved4b · Q5_0 · Vision — asked, failed4b · Q5_0 · Counts in images — never asked4b · Q5_0 · Multiple images at once — never asked4b · Q5_0 · Spatial reasoning — never asked4b · Q5_0 · Writes code — proved4b · Q5_0 · Reasons about code — proved4b · Q5_0 · Synthesises sources — proved4b · Q5_0 · Follows instructions — proved4b · Q5_K_M · Tool calling — proved4b · Q5_K_M · Tool loop finished — proved4b · Q5_K_M · Faithful to tool results — proved4b · Q5_K_M · Chained tool calls — proved4b · Q5_K_M · Structured output — proved4b · Q5_K_M · Sustained output — proved4b · Q5_K_M · Extended thinking — proved4b · Q5_K_M · Vision — asked, failed4b · Q5_K_M · Counts in images — never asked4b · Q5_K_M · Multiple images at once — never asked4b · Q5_K_M · Spatial reasoning — never asked4b · Q5_K_M · Writes code — proved4b · Q5_K_M · Reasons about code — proved4b · Q5_K_M · Synthesises sources — proved4b · Q5_K_M · Follows instructions — proved4b · Q8_0 · Tool calling — proved4b · Q8_0 · Tool loop finished — asked, failed4b · Q8_0 · Faithful to tool results — proved4b · Q8_0 · Chained tool calls — proved4b · Q8_0 · Structured output — proved4b · Q8_0 · Sustained output — proved4b · Q8_0 · Extended thinking — proved4b · Q8_0 · Vision — asked, failed4b · Q8_0 · Counts in images — never asked4b · Q8_0 · Multiple images at once — never asked4b · Q8_0 · Spatial reasoning — never asked4b · Q8_0 · Writes code — proved4b · Q8_0 · Reasons about code — proved4b · Q8_0 · Synthesises sources — proved4b · Q8_0 · Follows instructions — proved14b · Q8_0 · Tool calling — proved14b · Q8_0 · Tool loop finished — asked, failed14b · Q8_0 · Faithful to tool results — proved14b · Q8_0 · Chained tool calls — proved14b · Q8_0 · Structured output — proved14b · Q8_0 · Sustained output — proved14b · Q8_0 · Extended thinking — proved14b · Q8_0 · Vision — asked, failed14b · Q8_0 · Counts in images — never asked14b · Q8_0 · Multiple images at once — never asked14b · Q8_0 · Spatial reasoning — never asked14b · Q8_0 · Writes code — proved14b · Q8_0 · Reasons about code — proved14b · Q8_0 · Synthesises sources — proved14b · Q8_0 · Follows instructions — proved4b · Q4_K_M · Tool calling — proved4b · Q4_K_M · Tool loop finished — asked, failed4b · Q4_K_M · Faithful to tool results — proved4b · Q4_K_M · Chained tool calls — proved4b · Q4_K_M · Structured output — proved4b · Q4_K_M · Sustained output — proved4b · Q4_K_M · Extended thinking — proved4b · Q4_K_M · Vision — asked, failed4b · Q4_K_M · Counts in images — never asked4b · Q4_K_M · Multiple images at once — never asked4b · Q4_K_M · Spatial reasoning — never asked4b · Q4_K_M · Writes code — proved4b · Q4_K_M · Reasons about code — proved4b · Q4_K_M · Synthesises sources — proved4b · Q4_K_M · Follows instructions — proved4b · BF16 · Tool calling — proved4b · BF16 · Tool loop finished — asked, failed4b · BF16 · Faithful to tool results — proved4b · BF16 · Chained tool calls — proved4b · BF16 · Structured output — proved4b · BF16 · Sustained output — proved4b · BF16 · Extended thinking — proved4b · BF16 · Vision — never asked4b · BF16 · Counts in images — never asked4b · BF16 · Multiple images at once — never asked4b · BF16 · Spatial reasoning — never asked4b · BF16 · Writes code — proved4b · BF16 · Reasons about code — proved4b · BF16 · Synthesises sources — proved4b · BF16 · Follows instructions — proved4b · Q4_0 · Tool calling — proved4b · Q4_0 · Tool loop finished — proved4b · Q4_0 · Faithful to tool results — proved4b · Q4_0 · Chained tool calls — asked, failed4b · Q4_0 · Structured output — proved4b · Q4_0 · Sustained output — proved4b · Q4_0 · Extended thinking — asked, failed4b · Q4_0 · Vision — proved4b · Q4_0 · Counts in images — proved4b · Q4_0 · Multiple images at once — asked, failed4b · Q4_0 · Spatial reasoning — proved4b · Q4_0 · Writes code — proved4b · Q4_0 · Reasons about code — asked, failed4b · Q4_0 · Synthesises sources — asked, failed4b · Q4_0 · Follows instructions — provedvl-8b~prithivmlmods-abliterated · Q2_K · Tool calling — provedvl-8b~prithivmlmods-abliterated · Q2_K · Tool loop finished — provedvl-8b~prithivmlmods-abliterated · Q2_K · Faithful to tool results — asked, failedvl-8b~prithivmlmods-abliterated · Q2_K · Chained tool calls — never askedvl-8b~prithivmlmods-abliterated · Q2_K · Structured output — provedvl-8b~prithivmlmods-abliterated · Q2_K · Sustained output — provedvl-8b~prithivmlmods-abliterated · Q2_K · Extended thinking — asked, failedvl-8b~prithivmlmods-abliterated · Q2_K · Vision — provedvl-8b~prithivmlmods-abliterated · Q2_K · Counts in images — provedvl-8b~prithivmlmods-abliterated · Q2_K · Multiple images at once — asked, failedvl-8b~prithivmlmods-abliterated · Q2_K · Spatial reasoning — provedvl-8b~prithivmlmods-abliterated · Q2_K · Writes code — provedvl-8b~prithivmlmods-abliterated · Q2_K · Reasons about code — asked, failedvl-8b~prithivmlmods-abliterated · Q2_K · Synthesises sources — asked, failedvl-8b~prithivmlmods-abliterated · Q2_K · Follows instructions — proved4b · Q4_1 · unsloth · Tool calling — proved4b · Q4_1 · unsloth · Tool loop finished — proved4b · Q4_1 · unsloth · Faithful to tool results — proved4b · Q4_1 · unsloth · Chained tool calls — asked, failed4b · Q4_1 · unsloth · Structured output — proved4b · Q4_1 · unsloth · Sustained output — proved4b · Q4_1 · unsloth · Extended thinking — asked, failed4b · Q4_1 · unsloth · Vision — proved4b · Q4_1 · unsloth · Counts in images — proved4b · Q4_1 · unsloth · Multiple images at once — asked, failed4b · Q4_1 · unsloth · Spatial reasoning — proved4b · Q4_1 · unsloth · Writes code — proved4b · Q4_1 · unsloth · Reasons about code — asked, failed4b · Q4_1 · unsloth · Synthesises sources — asked, failed4b · Q4_1 · unsloth · Follows instructions — proved0.6b · Q4_K_M · Tool calling — proved0.6b · Q4_K_M · Tool loop finished — asked, failed0.6b · Q4_K_M · Faithful to tool results — proved0.6b · Q4_K_M · Chained tool calls — proved0.6b · Q4_K_M · Structured output — proved0.6b · Q4_K_M · Sustained output — proved0.6b · Q4_K_M · Extended thinking — proved0.6b · Q4_K_M · Vision — never asked0.6b · Q4_K_M · Counts in images — never asked0.6b · Q4_K_M · Multiple images at once — never asked0.6b · Q4_K_M · Spatial reasoning — never asked0.6b · Q4_K_M · Writes code — proved0.6b · Q4_K_M · Reasons about code — proved0.6b · Q4_K_M · Synthesises sources — proved0.6b · Q4_K_M · Follows instructions — proved0.6b · Q8_0 · Tool calling — proved0.6b · Q8_0 · Tool loop finished — asked, failed0.6b · Q8_0 · Faithful to tool results — asked, failed0.6b · Q8_0 · Chained tool calls — never asked0.6b · Q8_0 · Structured output — proved0.6b · Q8_0 · Sustained output — proved0.6b · Q8_0 · Extended thinking — proved0.6b · Q8_0 · Vision — asked, failed0.6b · Q8_0 · Counts in images — never asked0.6b · Q8_0 · Multiple images at once — never asked0.6b · Q8_0 · Spatial reasoning — never asked0.6b · Q8_0 · Writes code — proved0.6b · Q8_0 · Reasons about code — proved0.6b · Q8_0 · Synthesises sources — proved0.6b · Q8_0 · Follows instructions — proved4b · IQ4_XS · Tool calling — proved4b · IQ4_XS · Tool loop finished — asked, failed4b · IQ4_XS · Faithful to tool results — proved4b · IQ4_XS · Chained tool calls — asked, failed4b · IQ4_XS · Structured output — proved4b · IQ4_XS · Sustained output — never asked4b · IQ4_XS · Extended thinking — asked, failed4b · IQ4_XS · Vision — proved4b · IQ4_XS · Counts in images — proved4b · IQ4_XS · Multiple images at once — asked, failed4b · IQ4_XS · Spatial reasoning — proved4b · IQ4_XS · Writes code — proved4b · IQ4_XS · Reasons about code — asked, failed4b · IQ4_XS · Synthesises sources — never asked4b · IQ4_XS · Follows instructions — proved0.6b · Q2_K · Tool calling — never asked0.6b · Q2_K · Tool loop finished — never asked0.6b · Q2_K · Faithful to tool results — never asked0.6b · Q2_K · Chained tool calls — never asked0.6b · Q2_K · Structured output — never asked0.6b · Q2_K · Sustained output — never asked0.6b · Q2_K · Extended thinking — never asked0.6b · Q2_K · Vision — never asked0.6b · Q2_K · Counts in images — never asked0.6b · Q2_K · Multiple images at once — never asked0.6b · Q2_K · Spatial reasoning — never asked0.6b · Q2_K · Writes code — never asked0.6b · Q2_K · Reasons about code — never asked0.6b · Q2_K · Synthesises sources — never asked0.6b · Q2_K · Follows instructions — never asked4b · IQ1_S · Tool calling — asked, failed4b · IQ1_S · Tool loop finished — never asked4b · IQ1_S · Faithful to tool results — never asked4b · IQ1_S · Chained tool calls — never asked4b · IQ1_S · Structured output — asked, failed4b · IQ1_S · Sustained output — asked, failed4b · IQ1_S · Extended thinking — asked, failed4b · IQ1_S · Vision — never asked4b · IQ1_S · Counts in images — never asked4b · IQ1_S · Multiple images at once — never asked4b · IQ1_S · Spatial reasoning — never asked4b · IQ1_S · Writes code — asked, failed4b · IQ1_S · Reasons about code — asked, failed4b · IQ1_S · Synthesises sources — asked, failed4b · IQ1_S · Follows instructions — asked, failed4b · Q4_1 · unsloth · Tool calling — never asked4b · Q4_1 · unsloth · Tool loop finished — never asked4b · Q4_1 · unsloth · Faithful to tool results — never asked4b · Q4_1 · unsloth · Chained tool calls — never asked4b · Q4_1 · unsloth · Structured output — never asked4b · Q4_1 · unsloth · Sustained output — never asked4b · Q4_1 · unsloth · Extended thinking — never asked4b · Q4_1 · unsloth · Vision — never asked4b · Q4_1 · unsloth · Counts in images — never asked4b · Q4_1 · unsloth · Multiple images at once — never asked4b · Q4_1 · unsloth · Spatial reasoning — never asked4b · Q4_1 · unsloth · Writes code — never asked4b · Q4_1 · unsloth · Reasons about code — never asked4b · Q4_1 · unsloth · Synthesises sources — never asked4b · Q4_1 · unsloth · Follows instructions — never asked0.6b · Q3_K_M · Tool calling — proved0.6b · Q3_K_M · Tool loop finished — proved0.6b · Q3_K_M · Faithful to tool results — proved0.6b · Q3_K_M · Chained tool calls — asked, failed0.6b · Q3_K_M · Structured output — proved0.6b · Q3_K_M · Sustained output — proved0.6b · Q3_K_M · Extended thinking — proved0.6b · Q3_K_M · Vision — never asked0.6b · Q3_K_M · Counts in images — never asked0.6b · Q3_K_M · Multiple images at once — never asked0.6b · Q3_K_M · Spatial reasoning — never asked0.6b · Q3_K_M · Writes code — proved0.6b · Q3_K_M · Reasons about code — asked, failed0.6b · Q3_K_M · Synthesises sources — proved0.6b · Q3_K_M · Follows instructions — asked, failed0.6b · Q5_K_M · Tool calling — never asked0.6b · Q5_K_M · Tool loop finished — never asked0.6b · Q5_K_M · Faithful to tool results — never asked0.6b · Q5_K_M · Chained tool calls — never asked0.6b · Q5_K_M · Structured output — never asked0.6b · Q5_K_M · Sustained output — never asked0.6b · Q5_K_M · Extended thinking — never asked0.6b · Q5_K_M · Vision — never asked0.6b · Q5_K_M · Counts in images — never asked0.6b · Q5_K_M · Multiple images at once — never asked0.6b · Q5_K_M · Spatial reasoning — never asked0.6b · Q5_K_M · Writes code — never asked0.6b · Q5_K_M · Reasons about code — never asked0.6b · Q5_K_M · Synthesises sources — never asked0.6b · Q5_K_M · Follows instructions — never asked1.7b · Q8_0 · Tool calling — never asked1.7b · Q8_0 · Tool loop finished — never asked1.7b · Q8_0 · Faithful to tool results — never asked1.7b · Q8_0 · Chained tool calls — never asked1.7b · Q8_0 · Structured output — never asked1.7b · Q8_0 · Sustained output — never asked1.7b · Q8_0 · Extended thinking — never asked1.7b · Q8_0 · Vision — never asked1.7b · Q8_0 · Counts in images — never asked1.7b · Q8_0 · Multiple images at once — never asked1.7b · Q8_0 · Spatial reasoning — never asked1.7b · Q8_0 · Writes code — never asked1.7b · Q8_0 · Reasons about code — never asked1.7b · Q8_0 · Synthesises sources — never asked1.7b · Q8_0 · Follows instructions — never asked4b · Q2_K · Tool calling — proved4b · Q2_K · Tool loop finished — asked, failed4b · Q2_K · Faithful to tool results — proved4b · Q2_K · Chained tool calls — proved4b · Q2_K · Structured output — proved4b · Q2_K · Sustained output — proved4b · Q2_K · Extended thinking — asked, failed4b · Q2_K · Vision — proved4b · Q2_K · Counts in images — proved4b · Q2_K · Multiple images at once — asked, failed4b · Q2_K · Spatial reasoning — proved4b · Q2_K · Writes code — proved4b · Q2_K · Reasons about code — asked, failed4b · Q2_K · Synthesises sources — proved4b · Q2_K · Follows instructions — proved0.6b · Tool calling — never asked0.6b · Tool loop finished — never asked0.6b · Faithful to tool results — never asked0.6b · Chained tool calls — never asked0.6b · Structured output — never asked0.6b · Sustained output — never asked0.6b · Extended thinking — never asked0.6b · Vision — never asked0.6b · Counts in images — never asked0.6b · Multiple images at once — never asked0.6b · Spatial reasoning — never asked0.6b · Writes code — never asked0.6b · Reasons about code — never asked0.6b · Synthesises sources — never asked0.6b · Follows instructions — never asked
Tool calling
Tool loop finished
Faithful to tool results
Chained tool calls
Structured output
Sustained output
Extended thinking
Vision
Counts in images
Multiple images at once
Spatial reasoning
Writes code
Reasons about code
Synthesises sources
Follows instructions
Qwen 3 VL-4B1 build measured →
hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:F16vl-4b · F16 · Qwen · ollama · 1 machine · 4.02B
1,31,072recalled
—served
12.2best tok/s
recalled 128k⌐ declared 256k · unsignedadvertises 2× what it proved
Qwen 3 VL-8B1 build measured →
hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:F16vl-8b · F16 · Qwen · ollama · 1 machine · 8.19B
1,31,072recalled
—served
3.8best tok/s
recalled 128k⌐ declared 256k · unsignedadvertises 2× what it proved
30b-a3b1 build measured →
hf.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF:Q4_K_M30b-a3b · Q4_K_M · unsloth · ollama · 1 machine · 30.5B
1,31,072recalled
—served
26.7best tok/s
recalled 128k⌐ declared 256k · unsignedadvertises 2× what it proved
Qwen 3 1.7B2 builds measured →
Qwen/Qwen3-1.7B1.7b · qwen · vllm · 1 machine
65,536recalled
—served
—best tok/s
hf.co/Qwen/Qwen3-1.7B-GGUF:Q8_01.7b · Q8_0 · Qwen · ollama · 1 machine
—recalled
—served
5.9best tok/s
Qwen 3 14B2 builds measured →
Qwen/Qwen3-14B-AWQ14b · qwen · vllm · 1 machine
65,536recalled
—served
—best tok/s
hf.co/Qwen/Qwen3-14B-GGUF:Q8_014b · Q8_0 · Qwen · ollama · 1 machine · 14.8B
32,768recalled
—served
15.8best tok/s
recalled 32k⌐ declared 40k · unsigned
Qwen 3 4B13 builds measured →
Qwen/Qwen3-4B4b · qwen · vllm · 1 machine
65,536recalled
—served
—best tok/s
hf.co/unsloth/Qwen3-4B-GGUF:IQ3_XXS4b · IQ3_XXS · unsloth · ollama · 4 machines · 4.02B
32,768recalled
—served
62.9best tok/s
53.3 tok/s · unnamed machine47.8 tok/s · unnamed machine4.6 tok/s · unnamed machine0.2 tok/s · unnamed machine
0.2 → 53.3 tok/s across 4 machines · ×242.0 spreadunnamed machine
recalled 32k⌐ declared 40k · unsigned
hf.co/Qwen/Qwen3-4B-GGUF:Q5_04b · Q5_0 · Qwen · ollama · 3 machines · 4.02B
32,768recalled
—served
48.5best tok/s
48.0 tok/s · unnamed machine22.7 tok/s · unnamed machine11.2 tok/s · unnamed machine
11.2 → 48.0 tok/s across 3 machines · ×4.3 spreadunnamed machine
recalled 32k⌐ declared 40k · unsigned
hf.co/Qwen/Qwen3-4B-GGUF:Q5_K_M4b · Q5_K_M · Qwen · ollama · 2 machines · 4.02B
32,768recalled
—served
47.0best tok/s
47.0 tok/s · unnamed machine35.4 tok/s · unnamed machine
35.4 → 47.0 tok/s across 2 machines · ×1.3 spreadunnamed machine
recalled 32k⌐ declared 40k · unsigned
hf.co/Qwen/Qwen3-4B-GGUF:Q8_04b · Q8_0 · Qwen · ollama · 2 machines · 4.02B
32,768recalled
—served
38.9best tok/s
38.9 tok/s · unnamed machine17.1 tok/s · unnamed machine
17.1 → 38.9 tok/s across 2 machines · ×2.3 spreadunnamed machine
recalled 32k⌐ declared 40k · unsigned
hf.co/Qwen/Qwen3-4B-GGUF:Q4_K_M4b · Q4_K_M · Qwen · ollama · 2 machines · 4.02B
16,384recalled
—served
140.7best tok/s
recalled 16k⌐ declared 40k · unsignedadvertises 3× what it proved
hf.co/unsloth/Qwen3-4B-GGUF:BF164b · BF16 · unsloth · ollama · 1 machine · 4.02B
16,384recalled
—served
80.8best tok/s
recalled 16k⌐ declared 40k · unsignedadvertises 3× what it proved
hf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q4_04b · Q4_0 · unsloth · ollama · 4 machines · 4.02B
4,096recalled
—served
40.9best tok/s
40.9 tok/s · unnamed machine17.3 tok/s · unnamed machine17.3 tok/s · unnamed machine8.4 tok/s · unnamed machine
8.4 → 40.9 tok/s across 4 machines · ×4.9 spreadunnamed machine
recalled 4k⌐ declared 256k · unsignedadvertises 64× what it proved
hf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q4_14b · Q4_1 · unsloth · ollama · 2 machines
4,096recalled
—served
37.0best tok/s
37.0 tok/s · unnamed machine3.1 tok/s · unnamed machine
3.1 → 37.0 tok/s across 2 machines · ×11.9 spreadunnamed machine
hf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:IQ4_XS4b · IQ4_XS · unsloth · ollama · 1 machine · 4.02B
4,096recalled
—served
6.4best tok/s
recalled 4k⌐ declared 256k · unsignedadvertises 64× what it proved
hf.co/unsloth/Qwen3-4B-GGUF:IQ1_S4b · IQ1_S · unsloth · ollama · 3 machines · 4.02B
—recalled
—served
90.4best tok/s
90.4 tok/s · unnamed machine61.3 tok/s · unnamed machine14.4 tok/s · unnamed machine
14.4 → 90.4 tok/s across 3 machines · ×6.3 spreadunnamed machine
hf.co/unsloth/Qwen3-4B-GGUF:Q4_14b · Q4_1 · unsloth · ollama · 2 machines
—recalled
—served
19.4best tok/s
hf.co/unsloth/Qwen3-VL-4B-Instruct-GGUF:Q2_K4b · Q2_K · unsloth · ollama · 1 machine
—recalled
—served
11.9best tok/s
Qwen 3 8B1 build measured →
Qwen/Qwen3-8B8b · qwen · vllm · 1 machine
65,536recalled
—served
—best tok/s
Qwen3-VL-8B-Instruct-abliterated-v11 build measured →
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q2_Kvl-8b~prithivmlmods-abliterated · Q2_K · mradermacher · ollama · 3 machines
4,096recalled
—served
22.4best tok/s
22.4 tok/s · unnamed machine3.1 tok/s · unnamed machine2.3 tok/s · unnamed machine
2.3 → 22.4 tok/s across 3 machines · ×9.6 spreadunnamed machine
Qwen 3 0.6B6 builds measured →
hf.co/MaziyarPanahi/Qwen3-0.6B-GGUF:Q4_K_M0.6b · Q4_K_M · maziyarpanahi · ollama · 1 machine · 752M
4,096recalled
—served
37.4best tok/s
recalled 4k⌐ declared 40k · unsignedadvertises 10× what it proved
hf.co/Qwen/Qwen3-0.6B-GGUF:Q8_00.6b · Q8_0 · Qwen · ollama · 1 machine · 596M
4,096recalled
—served
15.9best tok/s
recalled 4k⌐ declared 40k · unsignedadvertises 10× what it proved
hf.co/MaziyarPanahi/Qwen3-0.6B-GGUF:Q2_K0.6b · Q2_K · maziyarpanahi · 3 machines
—recalled
—served
302.3best tok/s
302.3 tok/s · unnamed machine16.0 tok/s · unnamed machine7.1 tok/s · unnamed machine
7.1 → 302.3 tok/s across 3 machines · ×42.7 spreadunnamed machine
hf.co/MaziyarPanahi/Qwen3-0.6B-GGUF:Q3_K_M0.6b · Q3_K_M · maziyarpanahi · ollama · 1 machine · 752M
—recalled
—served
44.6best tok/s
hf.co/MaziyarPanahi/Qwen3-0.6B-GGUF:Q5_K_M0.6b · Q5_K_M · maziyarpanahi · 1 machine
—recalled
—served
23.9best tok/s
qwen3:0.6b0.6b · ollama · 1 machine
—recalled
—served
18.2best tok/s

Builds this network knows of

Declared, not measured — this is what the catalog says exists and what it would cost a machine to hold. Nothing here is signed, and none of it claims the build works.

how much memory does your machine have

41 of 54 builds fit a 16 GB machine — budgeting 70% of it, or 11.2 GiB, because the cache and the runtime want the rest.

  • hf.co/Qwen/Qwen3-0.6B-GGUF:Q8_0Q8_0ollamameasured herepublished by Qwen
    0.6 GiB
  • hf.co/Qwen/Qwen3-Embedding-0.6B-GGUF:Q8_0Q8_0ollamaderived by Qwen
    0.6 GiB
  • mlx-community/Qwen3-ASR-0.6B-8bit8BITmlxderived by mlx-community
    0.9 GiB
  • hf.co/unsloth/Qwen3-4B-GGUF:IQ1_SIQ1_Sollamameasured herequantized by unsloth
    1.0 GiB
  • hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:Q4_K_MQ4_K_Mollamapublished by Qwen
    1.0 GiB
  • hf.co/Qwen/Qwen3-Embedding-0.6B-GGUF:F16F16ollamaderived by Qwen
    1.1 GiB
  • hf.co/Serveurperso/Qwen3-TTS-GGUF:Q4_K_MQ4_K_Mollamaquantized by Serveurperso
    1.1 GiB
  • PrimeIntellect/Qwen3-0.6BBF16vllmderived by PrimeIntellect
    1.4 GiB
  • hf.co/unsloth/Qwen3-4B-GGUF:IQ3_XXSIQ3_XXSollamameasured herequantized by unsloth
    1.6 GiB
  • hf.co/Qwen/Qwen3-1.7B-GGUF:Q8_0Q8_0ollamameasured herepublished by Qwen
    1.7 GiB
  • hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:Q8_0Q8_0ollamapublished by Qwen
    1.7 GiB
  • hf.co/Serveurperso/Qwen3-TTS-GGUF:Q8_0Q8_0ollamaquantized by Serveurperso
    1.9 GiB
  • hf.co/Qwen/Qwen3-4B-GGUF:Q4_K_MQ4_K_Mollamameasured herepublished by Qwen
    2.3 GiB
  • hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:Q4_K_MQ4_K_Mollamapublished by Qwen
    2.3 GiB
  • hf.co/Qwen/Qwen3-4B-GGUF:Q5_0Q5_0ollamameasured herepublished by Qwen
    2.6 GiB
  • hf.co/Qwen/Qwen3-4B-GGUF:Q5_K_MQ5_K_Mollamameasured herepublished by Qwen
    2.7 GiB
  • lmstudio-community/Qwen3-VL-4B-Instruct-MLX-4bit4BITmlxquantized by lmstudio-community
    2.9 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q2_KQ2_Kollamameasured herederived by mradermacher
    3.1 GiB
  • hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:F16F16ollamapublished by Qwen
    3.2 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q3_K_SQ3_K_Sollamaderived by mradermacher
    3.5 GiB
  • hf.co/Serveurperso/Qwen3-TTS-GGUF:BF16BF16ollamaquantized by Serveurperso
    3.6 GiB
  • hf.co/Qwen/Qwen3-4B-GGUF:Q8_0Q8_0ollamameasured herepublished by Qwen
    4.0 GiB
  • hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:Q8_0Q8_0ollamapublished by Qwen
    4.0 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q3_K_LQ3_K_Lollamaderived by mradermacher
    4.1 GiB
  • lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit4BITmlxderived by lmstudio-community
    4.3 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q4_K_SQ4_K_Sollamaderived by mradermacher
    4.5 GiB
  • hf.co/Qwen/Qwen3-8B-GGUF:Q4_K_MQ4_K_Mollamapublished by Qwen
    4.7 GiB
  • hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:Q4_K_MQ4_K_Mollamapublished by Qwen
    4.7 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q4_K_MQ4_K_Mollamaderived by mradermacher
    4.7 GiB
  • hf.co/Qwen/Qwen3-8B-GGUF:Q5_0Q5_0ollamapublished by Qwen
    5.3 GiB
  • hf.co/Qwen/Qwen3-8B-GGUF:Q5_K_MQ5_K_Mollamapublished by Qwen
    5.4 GiB
  • hf.co/Serveurperso/Qwen3-TTS-GGUF:F32F32ollamaquantized by Serveurperso
    7.2 GiB
  • hf.co/unsloth/Qwen3-4B-GGUF:BF16BF16ollamameasured herequantized by unsloth
    7.5 GiB
  • hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:F16F16ollamameasured herepublished by Qwen
    7.5 GiB
  • lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit8BITmlxderived by lmstudio-community
    8.1 GiB
  • hf.co/Qwen/Qwen3-8B-GGUF:Q8_0Q8_0ollamapublished by Qwen
    8.1 GiB
  • hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:Q8_0Q8_0ollamapublished by Qwen
    8.1 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q8_0Q8_0ollamaderived by mradermacher
    8.1 GiB
  • hf.co/Qwen/Qwen3-14B-GGUF:Q4_K_MQ4_K_Mollamapublished by Qwen
    8.4 GiB
  • hf.co/Qwen/Qwen3-14B-GGUF:Q5_0Q5_0ollamapublished by Qwen
    9.6 GiB
  • hf.co/Qwen/Qwen3-14B-GGUF:Q5_K_MQ5_K_Mollamapublished by Qwen
    9.8 GiB
  • hf.co/Qwen/Qwen3-14B-GGUF:Q8_0Q8_0ollamameasured herepublished by Qwen
    14.6 GiB
  • hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:F16F16ollamameasured herepublished by Qwen
    15.3 GiB
  • hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:F16F16ollamaderived by mradermacher
    15.3 GiB
  • lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit4BITmlxquantized by lmstudio-community
    16.0 GiB
  • QuantTrio/Qwen3-VL-30B-A3B-Instruct-AWQAWQvllmquantized by QuantTrio
    16.7 GiB
  • cyankiwi/Qwen3-Coder-30B-A3B-Instruct-AWQ-4bitW4A16vllmquantized by cyankiwi
    16.9 GiB
  • lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit5BITmlxquantized by lmstudio-community
    19.5 GiB
  • lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit8BITmlxquantized by lmstudio-community
    30.2 GiB
  • lmstudio-community/Qwen3-Coder-Next-MLX-4bit4BITmlxquantized by lmstudio-community
    41.8 GiB
  • lmstudio-community/Qwen3-Coder-Next-MLX-6bit6BITmlxquantized by lmstudio-community
    60.3 GiB
  • Qwen/Qwen3-Coder-Next-FP8FP8vllmpublished by Qwen
    74.9 GiB
  • lmstudio-community/Qwen3-Coder-Next-MLX-8bit8BITmlxquantized by lmstudio-community
    78.8 GiB
  • Qwen/Qwen3-Coder-NextBF16vllmpublished by Qwen
    148.4 GiB

Sizes are what the catalog declares, not what a machine measured — fitting on disk is necessary and not sufficient. Nobody here has run these on your machine, and this narrows the candidates rather than certifying one.

the same builds, by member
Qwen 3 0.6B0.6B
hf.co/Qwen/Qwen3-0.6B-GGUF:Q8_0Q8_0ollama609 MiB on diskpublished by Qwen
Qwen3-0.6B
PrimeIntellect/Qwen3-0.6BBF16vllm1.4 GiB on diskderived by PrimeIntellect
Qwen3-Embedding-0.6B-GGUF
hf.co/Qwen/Qwen3-Embedding-0.6B-GGUF:Q8_0Q8_0ollama609 MiB on diskderived by Qwen
hf.co/Qwen/Qwen3-Embedding-0.6B-GGUF:F16F16ollama1.1 GiB on diskderived by Qwen
Qwen 3 1.7B
hf.co/Qwen/Qwen3-1.7B-GGUF:Q8_0Q8_0ollama1.7 GiB on diskpublished by Qwen
Qwen 3 14B
hf.co/Qwen/Qwen3-14B-GGUF:Q4_K_MQ4_K_Mollama8.4 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-14B-GGUF:Q5_0Q5_0ollama9.6 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-14B-GGUF:Q5_K_MQ5_K_Mollama9.8 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-14B-GGUF:Q8_0Q8_0ollama14.6 GiB on diskpublished by Qwen
Qwen 3 4B
hf.co/unsloth/Qwen3-4B-GGUF:IQ1_SIQ1_Sollama1.0 GiB on diskquantized by unsloth
hf.co/unsloth/Qwen3-4B-GGUF:IQ3_XXSIQ3_XXSollama1.6 GiB on diskquantized by unsloth
hf.co/Qwen/Qwen3-4B-GGUF:Q4_K_MQ4_K_Mollama2.3 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-4B-GGUF:Q5_0Q5_0ollama2.6 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-4B-GGUF:Q5_K_MQ5_K_Mollama2.7 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-4B-GGUF:Q8_0Q8_0ollama4.0 GiB on diskpublished by Qwen
hf.co/unsloth/Qwen3-4B-GGUF:BF16BF16ollama7.5 GiB on diskquantized by unsloth
Qwen 3 8B
hf.co/Qwen/Qwen3-8B-GGUF:Q4_K_MQ4_K_Mollama4.7 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-8B-GGUF:Q5_0Q5_0ollama5.3 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-8B-GGUF:Q5_K_MQ5_K_Mollama5.4 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-8B-GGUF:Q8_0Q8_0ollama8.1 GiB on diskpublished by Qwen
DeepSeek-R1-0528-Qwen3-8B
lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bit4BITmlx4.3 GiB on diskderived by lmstudio-community
lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bit8BITmlx8.1 GiB on diskderived by lmstudio-community
Qwen3-ASR-0.6B-8bit
mlx-community/Qwen3-ASR-0.6B-8bit8BITmlx959 MiB on diskderived by mlx-community
Qwen 3 CODER-30B-A3B
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bit4BITmlx16.0 GiB on diskquantized by lmstudio-community
cyankiwi/Qwen3-Coder-30B-A3B-Instruct-AWQ-4bitW4A16vllm16.9 GiB on diskquantized by cyankiwi
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bit5BITmlx19.5 GiB on diskquantized by lmstudio-community
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bit8BITmlx30.2 GiB on diskquantized by lmstudio-community
Qwen 3 CODER-NEXT
lmstudio-community/Qwen3-Coder-Next-MLX-4bit4BITmlx41.8 GiB on diskquantized by lmstudio-community
lmstudio-community/Qwen3-Coder-Next-MLX-6bit6BITmlx60.3 GiB on diskquantized by lmstudio-community
Qwen/Qwen3-Coder-Next-FP8FP8vllm74.9 GiB on diskpublished by Qwen
lmstudio-community/Qwen3-Coder-Next-MLX-8bit8BITmlx78.8 GiB on diskquantized by lmstudio-community
Qwen/Qwen3-Coder-NextBF16vllm148.4 GiB on diskpublished by Qwen
Qwen 3 TTS-0.6B-BASE
hf.co/Serveurperso/Qwen3-TTS-GGUF:Q4_K_MQ4_K_Mollama1.1 GiB on diskquantized by Serveurperso
hf.co/Serveurperso/Qwen3-TTS-GGUF:Q8_0Q8_0ollama1.9 GiB on diskquantized by Serveurperso
hf.co/Serveurperso/Qwen3-TTS-GGUF:BF16BF16ollama3.6 GiB on diskquantized by Serveurperso
hf.co/Serveurperso/Qwen3-TTS-GGUF:F32F32ollama7.2 GiB on diskquantized by Serveurperso
Qwen 3 VL-2B
hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:Q4_K_MQ4_K_Mollama1.0 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:Q8_0Q8_0ollama1.7 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:F16F16ollama3.2 GiB on diskpublished by Qwen
Qwen 3 VL-30B-A3B
QuantTrio/Qwen3-VL-30B-A3B-Instruct-AWQAWQvllm16.7 GiB on diskquantized by QuantTrio
Qwen 3 VL-4B
hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:Q4_K_MQ4_K_Mollama2.3 GiB on diskpublished by Qwen
lmstudio-community/Qwen3-VL-4B-Instruct-MLX-4bit4BITmlx2.9 GiB on diskquantized by lmstudio-community
hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:Q8_0Q8_0ollama4.0 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-VL-4B-Instruct-GGUF:F16F16ollama7.5 GiB on diskpublished by Qwen
Qwen 3 VL-8B
hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:Q4_K_MQ4_K_Mollama4.7 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:Q8_0Q8_0ollama8.1 GiB on diskpublished by Qwen
hf.co/Qwen/Qwen3-VL-8B-Instruct-GGUF:F16F16ollama15.3 GiB on diskpublished by Qwen
Qwen3-VL-8B-Instruct-abliterated-v1
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q2_KQ2_Kollama3.1 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q3_K_SQ3_K_Sollama3.5 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q3_K_LQ3_K_Lollama4.1 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q4_K_SQ4_K_Sollama4.5 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q4_K_MQ4_K_Mollama4.7 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:Q8_0Q8_0ollama8.1 GiB on diskderived by mradermacher
hf.co/mradermacher/Qwen3-VL-8B-Instruct-abliterated-GGUF:F16F16ollama15.3 GiB on diskderived by mradermacher
ask your own gram

Your concierge can answer this from the network rather than from this page. The ask carries the operator id, so it resolves to a capability and an engine — not to a name somebody has to recognise.

What has this network measured about the Qwen 3 family (epn.family.qwen-3), and can my gram run any of it? See /inferences/qwen-3

open a gramx with this ask →epn.family.qwen-3