model family
Gemma 3 logo

Gemma 3

Published by google. 2 members measured on real machines, across 13 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.

13 of 47 builds measured · 34 still open
what each build proved
270m · Q8_0
270m · F16
1b · Q4_K_M
1b · Q8_0
1b · Q3_K_L
1b · Q6_K
1b · Q2_K
1b · Q3_K_S
270m · Q4_K_M
270m · IQ2_XXS
270m · IQ2_M
270m · IQ3_XXS
270m · Q4_0
270m · Q8_0 · Structured output — proved270m · Q8_0 · Sustained output — asked, failed270m · Q8_0 · Writes code — proved270m · Q8_0 · Reasons about code — asked, failed270m · Q8_0 · Follows instructions — asked, failed270m · F16 · Structured output — proved270m · F16 · Sustained output — asked, failed270m · F16 · Writes code — proved270m · F16 · Reasons about code — asked, failed270m · F16 · Follows instructions — asked, failed1b · Q4_K_M · Structured output — proved1b · Q4_K_M · Sustained output — proved1b · Q4_K_M · Writes code — proved1b · Q4_K_M · Reasons about code — asked, failed1b · Q4_K_M · Follows instructions — proved1b · Q8_0 · Structured output — proved1b · Q8_0 · Sustained output — proved1b · Q8_0 · Writes code — proved1b · Q8_0 · Reasons about code — asked, failed1b · Q8_0 · Follows instructions — proved1b · Q3_K_L · Structured output — proved1b · Q3_K_L · Sustained output — proved1b · Q3_K_L · Writes code — proved1b · Q3_K_L · Reasons about code — asked, failed1b · Q3_K_L · Follows instructions — proved1b · Q6_K · Structured output — proved1b · Q6_K · Sustained output — proved1b · Q6_K · Writes code — proved1b · Q6_K · Reasons about code — asked, failed1b · Q6_K · Follows instructions — proved1b · Q2_K · Structured output — proved1b · Q2_K · Sustained output — proved1b · Q2_K · Writes code — proved1b · Q2_K · Reasons about code — asked, failed1b · Q2_K · Follows instructions — proved1b · Q3_K_S · Structured output — proved1b · Q3_K_S · Sustained output — proved1b · Q3_K_S · Writes code — proved1b · Q3_K_S · Reasons about code — asked, failed1b · Q3_K_S · Follows instructions — proved270m · Q4_K_M · Structured output — proved270m · Q4_K_M · Sustained output — asked, failed270m · Q4_K_M · Writes code — proved270m · Q4_K_M · Reasons about code — proved270m · Q4_K_M · Follows instructions — asked, failed270m · IQ2_XXS · Structured output — proved270m · IQ2_XXS · Sustained output — asked, failed270m · IQ2_XXS · Writes code — asked, failed270m · IQ2_XXS · Reasons about code — asked, failed270m · IQ2_XXS · Follows instructions — asked, failed270m · IQ2_M · Structured output — proved270m · IQ2_M · Sustained output — asked, failed270m · IQ2_M · Writes code — proved270m · IQ2_M · Reasons about code — asked, failed270m · IQ2_M · Follows instructions — asked, failed270m · IQ3_XXS · Structured output — proved270m · IQ3_XXS · Sustained output — asked, failed270m · IQ3_XXS · Writes code — proved270m · IQ3_XXS · Reasons about code — asked, failed270m · IQ3_XXS · Follows instructions — asked, failed270m · Q4_0 · Structured output — proved270m · Q4_0 · Sustained output — proved270m · Q4_0 · Writes code — proved270m · Q4_0 · Reasons about code — asked, failed270m · Q4_0 · Follows instructions — asked, failed
Structured output
Sustained output
Writes code
Reasons about code
Follows instructions
Gemma 3 270M7 builds measured →
hf.co/unsloth/gemma-3-270m-it-GGUF:Q8_0270m · Q8_0 · unsloth · ollama · 1 machine · 268M
32Krecalled
—served
31.3best tok/s
recalled 32k⌐ declared 32k · unsigned
hf.co/unsloth/gemma-3-270m-it-GGUF:F16270m · F16 · unsloth · ollama · 1 machine · 268M
16Krecalled
—served
23.5best tok/s
recalled 16k⌐ declared 32k · unsignedadvertises 2× what it proved
hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_K_M270m · Q4_K_M · unsloth · ollama · 6 machines · 268M
—recalled
—served
291.8best tok/s
291.8 tok/s · unnamed machine286.1 tok/s · unnamed machine51.5 tok/s · x86 · CPU · Bengaluru41.4 tok/s · x86 · CPU · Hyderabad37.4 tok/s · unnamed machine27.3 tok/s · unnamed machine
27.3 → 291.8 tok/s across 6 machines · ×10.7 spreadunnamed machinex86 · CPU
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ2_XXS270m · IQ2_XXS · unsloth · ollama · 2 machines · 268M
—recalled
—served
34.5best tok/s
34.5 tok/s · unnamed machine16.9 tok/s · x86 · CPU · Hyderabad
16.9 → 34.5 tok/s across 2 machines · ×2.0 spreadunnamed machinex86 · CPU
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ2_M270m · IQ2_M · unsloth · ollama · 1 machine · 268M
—recalled
—served
34.8best tok/s
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ3_XXS270m · IQ3_XXS · unsloth · ollama · 1 machine · 268M
—recalled
—served
33.4best tok/s
hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_0270m · Q4_0 · unsloth · ollama · 1 machine · 268M
—recalled
—served
33.0best tok/s
Gemma 3 1B6 builds measured →
hf.co/ggml-org/gemma-3-1b-it-GGUF:Q4_K_M1b · Q4_K_M · ggml-org · ollama · 2 machines · 1,000M
8Krecalled
—served
14.0best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/ggml-org/gemma-3-1b-it-GGUF:Q8_01b · Q8_0 · ggml-org · ollama · 1 machine · 1,000M
8Krecalled
—served
9.7best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q3_K_L1b · Q3_K_L · lmstudio-community · ollama · 1 machine · 1,000M
8Krecalled
—served
13.1best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q6_K1b · Q6_K · lmstudio-community · ollama · 1 machine · 1,000M
8Krecalled
—served
10.8best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/MaziyarPanahi/gemma-3-1b-it-GGUF:Q2_K1b · Q2_K · maziyarpanahi · ollama · 1 machine · 1,000M
8Krecalled
—served
13.0best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved
hf.co/MaziyarPanahi/gemma-3-1b-it-GGUF:Q3_K_S1b · Q3_K_S · maziyarpanahi · ollama · 1 machine · 1,000M
8Krecalled
—served
13.6best tok/s
recalled 8k⌐ declared 32k · unsignedadvertises 4× what it proved

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

38 of 47 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/unsloth/gemma-3-270m-it-GGUF:IQ2_XXSIQ2_XXSollamameasured herequantized by unsloth
    0.2 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:IQ2_MIQ2_Mollamameasured herequantized by unsloth
    0.2 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:IQ3_XXSIQ3_XXSollamameasured herequantized by unsloth
    0.2 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_0Q4_0ollamameasured herequantized by unsloth
    0.2 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_K_MQ4_K_Mollamameasured herequantized by unsloth
    0.2 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:Q8_0Q8_0ollamameasured herequantized by unsloth
    0.3 GiB
  • lmstudio-community/gemma-3-270m-it-MLX-8bit8BITmlxquantized by lmstudio-community
    0.4 GiB
  • hf.co/unsloth/gemma-3-270m-it-GGUF:F16F16ollamameasured herequantized by unsloth
    0.5 GiB
  • hf.co/unsloth/gemma-3-1b-it-GGUF:IQ1_SIQ1_Sollamaquantized by unsloth
    0.5 GiB
  • hf.co/unsloth/gemma-3-1b-it-GGUF:IQ3_XXSIQ3_XXSollamaquantized by unsloth
    0.6 GiB
  • mlx-community/gemma-3-1b-it-qat-4bit4BITmlxquantized by mlx-community
    0.7 GiB
  • hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q3_K_LQ3_K_Lollamameasured herequantized by lmstudio-community
    0.7 GiB
  • hf.co/ggml-org/gemma-3-1b-it-GGUF:Q4_K_MQ4_K_Mollamameasured herequantized by ggml-org
    0.8 GiB
  • hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q6_KQ6_Kollamameasured herequantized by lmstudio-community
    0.9 GiB
  • Oscilla/gemma-3-1b-it-mlx-8Bit8BITmlxquantized by Oscilla
    1.0 GiB
  • hf.co/ggml-org/gemma-3-1b-it-GGUF:Q8_0Q8_0ollamameasured herequantized by ggml-org
    1.0 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:IQ1_SIQ1_Sollamaquantized by unsloth
    1.1 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:IQ2_XXSIQ2_XXSollamaquantized by unsloth
    1.3 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:IQ3_XXSIQ3_XXSollamaquantized by unsloth
    1.6 GiB
  • hf.co/ggml-org/gemma-3-1b-it-GGUF:F16F16ollamaquantized by ggml-org
    1.9 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:Q4_0Q4_0ollamaquantized by unsloth
    2.2 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:Q4_K_MQ4_K_Mollamaquantized by unsloth
    2.3 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ1_SIQ1_Sollamaquantized by unsloth
    2.8 GiB
  • hf.co/google/gemma-3-4b-it-qat-q4_0-gguf:Q4_0Q4_0ollamaderived by google
    2.9 GiB
  • mlx-community/gemma-3-4b-it-4bitBF16mlxquantized by mlx-community
    3.2 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ2_XXSIQ2_XXSollamaquantized by unsloth
    3.4 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:Q8_0Q8_0ollamaquantized by unsloth
    3.8 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ3_XXSIQ3_XXSollamaquantized by unsloth
    4.5 GiB
  • mlx-community/gemma-3-4b-it-8bitBF16mlxderived by mlx-community
    5.3 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:IQ1_SIQ1_Sollamaquantized by unsloth
    6.1 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q4_0Q4_0ollamaquantized by unsloth
    6.4 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q4_K_MQ4_K_Mollamaquantized by unsloth
    6.8 GiB
  • hf.co/unsloth/gemma-3-4b-it-GGUF:BF16BF16ollamaquantized by unsloth
    7.2 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:IQ2_XXSIQ2_XXSollamaquantized by unsloth
    7.3 GiB
  • mlx-community/gemma-3-12b-it-4bitBF16mlxderived by mlx-community
    7.5 GiB
  • abhishekchohan/gemma-3-12b-it-quantized-W4A16W4A16vllmquantized by abhishekchohan
    7.8 GiB
  • gaunernst/gemma-3-12b-it-int4-awqAWQvllmquantized by gaunernst
    8.5 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:IQ3_XXSIQ3_XXSollamaquantized by unsloth
    10.1 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q8_0Q8_0ollamaquantized by unsloth
    11.7 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:Q4_0Q4_0ollamaquantized by unsloth
    14.5 GiB
  • stephenwalker/gemma-3-27b-it-mlxBF16mlxderived by stephenwalker
    15.3 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:Q4_K_MQ4_K_Mollamaquantized by unsloth
    15.4 GiB
  • ISTA-DASLab/gemma-3-27b-it-GPTQ-4b-128gW4A16vllmderived by ISTA-DASLab
    15.7 GiB
  • gaunernst/gemma-3-27b-it-int4-awqAWQvllmquantized by gaunernst
    17.2 GiB
  • hf.co/unsloth/gemma-3-12b-it-qat-GGUF:BF16BF16ollamaquantized by unsloth
    21.9 GiB
  • hf.co/unsloth/gemma-3-27b-it-GGUF:Q8_0Q8_0ollamaquantized by unsloth
    26.7 GiB
  • unsloth/gemma-3-27b-it-GGUF:BF16BF16llamacppquantized by unsloth
    50.3 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
Gemma 3 12B
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ1_SIQ1_Sollama2.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ2_XXSIQ2_XXSollama3.4 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:IQ3_XXSIQ3_XXSollama4.5 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q4_0Q4_0ollama6.4 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q4_K_MQ4_K_Mollama6.8 GiB on diskquantized by unsloth
abhishekchohan/gemma-3-12b-it-quantized-W4A16W4A16vllm7.8 GiB on diskquantized by abhishekchohan
gaunernst/gemma-3-12b-it-int4-awqAWQvllm8.5 GiB on diskquantized by gaunernst
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:Q8_0Q8_0ollama11.7 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-12b-it-qat-GGUF:BF16BF16ollama21.9 GiB on diskquantized by unsloth
gemma-3-12b-it-4bit
mlx-community/gemma-3-12b-it-4bitBF16mlx7.5 GiB on diskderived by mlx-community
Gemma 3 1B
hf.co/unsloth/gemma-3-1b-it-GGUF:IQ1_SIQ1_Sollama531 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-1b-it-GGUF:IQ3_XXSIQ3_XXSollama564 MiB on diskquantized by unsloth
mlx-community/gemma-3-1b-it-qat-4bit4BITmlx698 MiB on diskquantized by mlx-community
hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q3_K_LQ3_K_Lollama716 MiB on diskquantized by lmstudio-community
hf.co/ggml-org/gemma-3-1b-it-GGUF:Q4_K_MQ4_K_Mollama768 MiB on diskquantized by ggml-org
hf.co/lmstudio-community/gemma-3-1b-it-GGUF:Q6_KQ6_Kollama964 MiB on diskquantized by lmstudio-community
Oscilla/gemma-3-1b-it-mlx-8Bit8BITmlx1013 MiB on diskquantized by Oscilla
hf.co/ggml-org/gemma-3-1b-it-GGUF:Q8_0Q8_0ollama1019 MiB on diskquantized by ggml-org
hf.co/ggml-org/gemma-3-1b-it-GGUF:F16F16ollama1.9 GiB on diskquantized by ggml-org
Gemma 3 270M
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ2_XXSIQ2_XXSollama171 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ2_MIQ2_Mollama174 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-270m-it-GGUF:IQ3_XXSIQ3_XXSollama175 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_0Q4_0ollama230 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-270m-it-GGUF:Q4_K_MQ4_K_Mollama241 MiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-270m-it-GGUF:Q8_0Q8_0ollama278 MiB on diskquantized by unsloth
lmstudio-community/gemma-3-270m-it-MLX-8bit8BITmlx441 MiB on diskquantized by lmstudio-community
hf.co/unsloth/gemma-3-270m-it-GGUF:F16F16ollama517 MiB on diskquantized by unsloth
Gemma 3 27B
hf.co/unsloth/gemma-3-27b-it-GGUF:IQ1_SIQ1_Sollama6.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-27b-it-GGUF:IQ2_XXSIQ2_XXSollama7.3 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-27b-it-GGUF:IQ3_XXSIQ3_XXSollama10.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-27b-it-GGUF:Q4_0Q4_0ollama14.5 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-27b-it-GGUF:Q4_K_MQ4_K_Mollama15.4 GiB on diskquantized by unsloth
gaunernst/gemma-3-27b-it-int4-awqAWQvllm17.2 GiB on diskquantized by gaunernst
hf.co/unsloth/gemma-3-27b-it-GGUF:Q8_0Q8_0ollama26.7 GiB on diskquantized by unsloth
unsloth/gemma-3-27b-it-GGUF:BF16BF16llamacpp50.3 GiB on diskquantized by unsloth
gemma-3-27b-it-GPTQ-4b-128g
ISTA-DASLab/gemma-3-27b-it-GPTQ-4b-128gW4A16vllm15.7 GiB on diskderived by ISTA-DASLab
gemma-3-27b-it-mlx
stephenwalker/gemma-3-27b-it-mlxBF16mlx15.3 GiB on diskderived by stephenwalker
Gemma 3 4B
hf.co/unsloth/gemma-3-4b-it-GGUF:IQ1_SIQ1_Sollama1.1 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-4b-it-GGUF:IQ2_XXSIQ2_XXSollama1.3 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-4b-it-GGUF:IQ3_XXSIQ3_XXSollama1.6 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-4b-it-GGUF:Q4_0Q4_0ollama2.2 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-4b-it-GGUF:Q4_K_MQ4_K_Mollama2.3 GiB on diskquantized by unsloth
mlx-community/gemma-3-4b-it-4bitBF16mlx3.2 GiB on diskquantized by mlx-community
hf.co/unsloth/gemma-3-4b-it-GGUF:Q8_0Q8_0ollama3.8 GiB on diskquantized by unsloth
hf.co/unsloth/gemma-3-4b-it-GGUF:BF16BF16ollama7.2 GiB on diskquantized by unsloth
gemma-3-4b-it-qat-q4_0-gguf
hf.co/google/gemma-3-4b-it-qat-q4_0-gguf:Q4_0Q4_0ollama2.9 GiB on diskderived by google
gemma-3-4b-it-8bit
mlx-community/gemma-3-4b-it-8bitBF16mlx5.3 GiB on diskderived by mlx-community
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 Gemma 3 family (epn.family.gemma-3), and can my gram run any of it? See /inferences/gemma-3

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