Inference · every build

Every build, with its own numbers.

A build is one file: a packager’s quantization of one model, run by one engine. Each row is measured on real machines and signed by the machine that measured it. Speeds are tokens a second; the fastest names its machine, the typical is the middle of every machine that ran it.

365builds measured
169models
76families
34machines

Gemma 4

Gemma 4 E2B12 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama4.4291RTX PRO 6000 Blackwell · 96 GB5413128K20 + 37 lang
IQ4_XS by unsloth · ollama3.4183GeForce RTX 3080 Ti · 12 GB793128K21 + 29 lang
Q4_K_M by google · ollama1.3170Windows, 16 CPUs · 16 GB1910128K20 + 37 lang
Q4_0 by unsloth · ollama2.685583128K21 + 10 lang
Q8_0 by unsloth · ollama5.7675.32128K21 + 22 lang
IQ4_NL by unsloth · ollama—553228K21 + 14 lang
Q3_K_M by unsloth · ollama—381438K19 + 10 lang
BF16 by unsloth · ollama15383.12128K21 + 20 lang
Q3_K_S by unsloth · ollama—14—14K19 + 14 lang
ollama by ollama · ollama—8.9—1—0
llamacpp by google · llamacpp—2.0—1—1
Q2_K_XL by unsloth · llamacpp—not timed—1—10 + 6 lang

Gemma 4 E4B8 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by google · ollama3.294—1128K21 + 14 lang
IQ4_XS by unsloth · ollama5.961—1128K21 + 10 lang
Q4_K_M by unsloth · ollama6.257—1128K20 + 10 lang
Q4_0 by unsloth · ollama6.055—1128K21 + 9 lang
Q8_0 by unsloth · ollama8.339—1128K19 + 14 lang
BF16 by unsloth · ollama1126—1128K21 + 13 lang
Q3_K_M by unsloth · ollama—20—14K21 + 9 lang
ollama by ollama · ollama—not timed—1—0

Gemma 4 E2B · bartowski7 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_0 by bartowski · ollama2.687—1128K21 + 10 lang
Q4_K_M by bartowski · ollama2.773—1128K21 + 10 lang
Q3_K_L by bartowski · ollama2.573—1128K21 + 10 lang
IQ2_M by bartowski · ollama2.271—1128K20 + 10 lang
IQ3_XS by bartowski · ollama2.362—1128K21 + 10 lang
Q8_0 by bartowski · ollama3.661—1128K21 + 10 lang
BF16 by bartowski · ollama1528—1128K21 + 10 lang

Gemma 4 12B5 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama14291.52128K21 + 15 lang
Q4_0 by google · ollama1428—1128K21 + 10 lang
IQ4_XS by unsloth · ollama7.223—1128K21 + 10 lang
Q8_0 by unsloth · ollama1619—1256K21 + 14 lang
BF16 by unsloth · ollama2611—1128K16 + 14 lang

Gemma 4 12B · yuxinlu14 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by yuxinlu1 · ollama8.026—1128K14 + 13 lang
Q6_K by yuxinlu1 · ollama1720—1128K14 + 10 lang
Q8_0 by yuxinlu1 · ollama1618—1128K14 + 10 lang
Q3_K_M by yuxinlu1 · ollama6.817—1128K13 + 6 lang

Gemma 4 26B-A4B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by unsloth · ollama23213.02128K21 + 21 lang

Gemma 4 31B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama203.4—1128K21 + 22 lang
Q4_0 by unsloth · ollama172.4—1128K21 + 10 lang

Phi 4

Phi 4 MINI21 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by maziyarpanahi · ollama5.8205GeForce RTX 3080 Ti · 12 GB—14K6 + 5 lang
Q2_K by maziyarpanahi · ollama—154Windows, 16 CPUs · 16 GB—1—4 + 1 lang
Q3_K_S by maziyarpanahi · ollama—135Windows, 16 CPUs · 16 GB2834K6 + 1 lang
Q3_K_L by maziyarpanahi · ollama22110Windows, 16 CPUs · 16 GB643128K5 + 1 lang
Q4_K_M by unsloth · ollama22823.15128K6 + 2 lang
Q4_K_S by maziyarpanahi · ollama—77Windows, 16 CPUs · 16 GB—1—6 + 2 lang
Q2_K by unsloth · ollama2.371—1—4 + 1 lang
Q5_K_M by unsloth · ollama2369—1128K6 + 4 lang
Q3_K_M by unsloth · ollama2267Linux, 4 CPUs · 31 GB—1128K4 + 2 lang
Q8_0 by unsloth · ollama2454—1128K6 + 2 lang
Q4_K_M by bartowski · ollama3.141—14K6 + 2 lang
Q8_0 by maziyarpanahi · ollama—39Windows, 16 CPUs · 16 GB—1—5 + 2 lang
IQ2_M by bartowski · ollama2.1371424K5 + 1 lang
IQ4_XS by bartowski · ollama2.836Mac, 10 CPUs · 16 GB—14K5 + 2 lang
Q4_0 by bartowski · ollama2.934Mac, 10 CPUs · 16 GB—14K6 + 1 lang
Q3_K_L by bartowski · ollama—32—18K6 + 3 lang
BF16 by unsloth · ollama26320.12128K5 + 3 lang
Q3_K_S by bartowski · ollama—28Mac, 10 CPUs · 16 GB—18K5 + 2 lang
IQ2_XS by mungert · ollama—8.5—1—4 + 1 lang
IQ3_XXS by bartowski · ollama2.30.7Mac, 10 CPUs · 16 GB—14K6 + 1 lang
BF16 by microsoft · vllm—not timed—14K6 + 1 lang

Phi 4 MINI-REASONING10 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XXS by unsloth · ollama10189Windows, 16 CPUs · 16 GB3464K7 + 2 lang
IQ1_S by unsloth · ollama—1895.69—0 + 1 lang
Q3_K_L by lmstudio-community · ollama—90Windows, 16 CPUs · 16 GB144—7 + 1 lang
Q2_K by unsloth · ollama—38243—7 + 1 lang
Q3_K_M by unsloth · ollama—32204—7 + 1 lang
Q8_0 by lmstudio-community · ollama4.722Mac, 10 CPUs · 16 GB—14K7 + 2 lang
Q4_K_M by lmstudio-community · ollama—7.3Mac, 8 CPUs · 8 GB6.02—7 + 1 lang
Q2_K_L by unsloth · ollama—6.86.83—5
Q4_K_M by unsloth · ollama—6.5Windows, 8 CPUs · 16 GB4.22—6
Q6_K by lmstudio-community · ollama—not timed—1—0

Phi 4 REASONING-PLUS3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama1725—132K5 + 9 lang
Q6_K by lmstudio-community · ollama—3.3Mac, 10 CPUs · 16 GB—1—6 + 3 lang
Q4_K_M by lmstudio-community · ollama—not timed—1—0

Phi 4 BASE2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by microsoft · ollama—3.6—1—0
vllm by redhatai · vllm—not timed—132K7 + 6 lang

FunctionGemma

FunctionGemma 270M7 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama0.4341L4 · 23 GB343—4 + 3 lang
IQ2_XXS by unsloth · ollama0.367Intel(R) Iris(R) Xe Graphics · 1 GB502—2
Q4_0 by unsloth · ollama0.337Linux, 4 CPUs · 31 GB—1—4 + 1 lang
Q8_0 by unsloth · ollama0.333Linux, 4 CPUs · 31 GB—1—2 + 2 lang
IQ3_XXS by unsloth · ollama0.233Linux, 4 CPUs · 31 GB—1—2
IQ2_M by unsloth · ollama0.233Linux, 4 CPUs · 31 GB182—2
BF16 by unsloth · ollama0.619Linux, 4 CPUs · 31 GB—1—2 + 3 lang

functiongemma-270m-it-GGUF6 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ4_NL by Mungert · ollama0.340—1—2 + 2 lang
Q3_K_S by Mungert · ollama0.234—1—2 + 1 lang
Q4_1 by Mungert · ollama0.234—1—2 + 3 lang
Q4_K_M by Mungert · ollama0.333—1—2 + 2 lang
Q8_0 by Mungert · ollama0.522—1—2 + 2 lang
BF16 by Mungert · ollama0.620—1—2 + 3 lang

functiongemma-270m-it-unsloth-bnb-4bit-ft-QAT-GGUF4 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mingxilei · ollama0.339Linux, 4 CPUs · 31 GB—1—2
Q5_K_M by mingxilei · ollama0.333—1—3
Q8_0 by mingxilei · ollama0.331Linux, 4 CPUs · 31 GB—1—3
F16 by mingxilei · ollama0.623Linux, 4 CPUs · 31 GB—1—4

functiongemma-270m-it-abominable-heresy7 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ4_XS by mradermacher · ollama0.339Linux, 4 CPUs · 31 GB—1—2 + 2 lang
Q3_K_S by mradermacher · ollama0.337—1—2
Q2_K by mradermacher · ollama0.335Linux, 4 CPUs · 31 GB—1—2
Q4_K_M by mradermacher · ollama0.334Linux, 4 CPUs · 31 GB—1—3 + 1 lang
Q4_K_S by mradermacher · ollama0.334Linux, 4 CPUs · 31 GB—1—3 + 1 lang
Q8_0 by mradermacher · ollama0.332Linux, 4 CPUs · 31 GB—1—2 + 1 lang
F16 by mradermacher · ollama0.623Linux, 4 CPUs · 31 GB—1—2 + 1 lang

functiongemma-270m-it1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by victor · ollama0.337—1—4

Home-FunctionGemma-270m7 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q3_K_S by mradermacher · ollama0.336Linux, 4 CPUs · 31 GB—1—0
IQ4_XS by mradermacher · ollama0.335Linux, 4 CPUs · 31 GB—1—0
Q2_K by mradermacher · ollama0.335Linux, 4 CPUs · 31 GB—1—0
Q4_K_M by mradermacher · ollama0.333Linux, 4 CPUs · 31 GB—1—0
Q4_K_S by mradermacher · ollama0.333Linux, 4 CPUs · 31 GB—1—0
Q8_0 by mradermacher · ollama0.332Linux, 4 CPUs · 31 GB—1—0
F16 by mradermacher · ollama0.623Linux, 4 CPUs · 31 GB—1—0

Qwen 3

Qwen 3 0.6B6 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q2_K by maziyarpanahi—351163—0
Q3_K_M by maziyarpanahi · ollama—45—1—13 + 1 lang
Q4_K_M by maziyarpanahi · ollama—37—14K14 + 1 lang
Q5_K_M by maziyarpanahi—24—1—0
build by ollama—18—1—0
Q8_0 by Qwen · ollama1.116—14K11 + 1 lang

Qwen 3 4B13 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by Qwen · ollama9.014150340K14 + 4 lang
IQ1_S by unsloth · ollama1.890Windows, 16 CPUs · 16 GB613—0
BF16 by unsloth · ollama1181Windows, 16 CPUs · 16 GB—116K14 + 2 lang
IQ3_XXS by unsloth · ollama126326432K14 + 2 lang
Q5_0 by Qwen · ollama7.94923332K14 + 3 lang
Q5_K_M by Qwen · ollama7.947Linux, 4 CPUs · 31 GB41232K16 + 2 lang
Q4_0 by unsloth · ollama—411744K15 + 6 lang
Q8_0 by Qwen · ollama9.339Linux, 4 CPUs · 31 GB28232K14 + 1 lang
Q4_1 by unsloth · ollama—372024K15 + 6 lang
Q4_1 by unsloth—19162—0
Q2_K by unsloth · ollama—12—1—15 + 4 lang
IQ4_XS by unsloth · ollama—6.4—14K12 + 6 lang
vllm by qwen · vllm—not timed—164K12 + 1 lang

30b-a3b · unsloth1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama2927—1128K6 + 9 lang

Qwen3-VL-8B-Instruct-abliterated-v11 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q2_K by mradermacher · ollama—222.634K12 + 4 lang

Qwen 3 14B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by Qwen · ollama2216—132K14 + 6 lang
vllm by qwen · vllm—not timed—164K12 + 7 lang

Qwen 3 VL-4B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
F16 by Qwen · ollama3012Linux, 4 CPUs · 31 GB—1128K14 + 6 lang

Qwen 3 VL-2B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by Qwen · ollama2.09.3Linux, 4 CPUs · 31 GB—14K13

Qwen 3 1.7B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by Qwen · ollama—5.9—1—0
vllm by qwen · vllm—not timed—164K12 + 1 lang

Qwen 3 VL-8B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
F16 by Qwen · ollama323.8—1128K14 + 7 lang

Qwen 3 8B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
vllm by qwen · vllm—not timed—164K12 + 4 lang

Llama 3.2

Llama 3.2 3B13 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ1_S by unsloth · ollama1.5225Windows, 16 CPUs · 16 GB1602—0 + 1 lang
Q4_0 by unsloth · ollama—205Windows, 16 CPUs · 16 GB—1—10 + 1 lang
Q4_K_M by unsloth · ollama17204Windows, 16 CPUs · 16 GB973128K12 + 7 lang
IQ3_XXS by unsloth · ollama—200Windows, 16 CPUs · 16 GB—1—11 + 4 lang
IQ2_XXS by unsloth · ollama—124Windows, 16 CPUs · 16 GB—1—5 + 1 lang
Q4_K_M by bartowski · ollama2299GeForce RTX 3080 Ti · 12 GB932128K12 + 5 lang
F16 by unsloth · ollama8.895Windows, 16 CPUs · 16 GB—116K11 + 5 lang
Q3_K_L by bartowski · ollama2279Linux, 4 CPUs · 31 GB—1128K12 + 2 lang
Q8_0 by unsloth · ollama—73Windows, 16 CPUs · 16 GB—1—12 + 3 lang
IQ3_M by bartowski · ollama2272Linux, 4 CPUs · 31 GB—1128K12 + 3 lang
Q8_0 by bartowski · ollama2463—1128K10 + 3 lang
IQ1_S by maziyarpanahi · ollama—46—1—3
F16 by bartowski · ollama2438—1128K11 + 3 lang

Llama 3.2 1B9 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by maziyarpanahi · ollama—113Mac, 10 CPUs · 16 GB6128K11 + 2 lang
IQ3_XS by maziyarpanahi · ollama—1081834K10 + 4 lang
IQ1_S by maziyarpanahi · ollama0.61058.64—0
Q4_K_M by bartowski · ollama1.010412416K12 + 6 lang
Q8_0 by ollama · ollama2.6509.61632K11 + 7 lang
Q3_K_L by bartowski · ollama0.910Linux, 4 CPUs · 31 GB—1—9 + 3 lang
Q8_0 by bartowski · ollama1.58.7Linux, 4 CPUs · 31 GB—14K11 + 2 lang
F16 by bartowski · ollama2.78.05.52—11 + 5 lang
IQ3_M by bartowski · ollama0.95.3Linux, 4 CPUs · 31 GB—1—11 + 1 lang

Llama-3.2-3B-Instruct-uncensored1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XS by bartowski · ollama—47machine not reported—18K9 + 1 lang

Llama-3.2-1B-Instruct-Q4_K_M-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by hugging-quants · ollama1.011—1—3 + 1 lang

EmbeddingGemma

EmbeddingGemma 300M10 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by unsloth · ollama0.7not timed—1—1
F32 by unsloth · ollama1.3not timed—2—1
Q2_K by gaianet · ollama0.3not timed—1—1
Q2_K by second-state · ollama0.3not timed—1—1
Q3_K_S by gaianet · ollama0.3not timed—1—1
Q3_K_S by second-state · ollama0.3not timed—1—1
Q4_0 by unsloth · ollama0.3not timed—1—1
Q4_K_M by second-state · ollama0.3not timed—4—1
Q4_K_M by gaianet · ollama0.3not timed—3—1
Q8_0 by ggml-org · ollama0.4not timed—1—1

embeddinggemma-300m-gguf4 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
F16 by lm-kit · ollama0.7not timed—1—1
Q4_K_M by lm-kit · ollama0.3not timed—1—1
Q5_K_M by lm-kit · ollama0.3not timed—1—1
Q8_0 by lm-kit · ollama0.4not timed—1—1

embeddinggemma-300m-GGUF-with-dense-modules2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q6_K by cduk · ollama0.3not timed—1—1
Q8_0 by cduk · ollama0.4not timed—1—1

Gemma 3

Gemma 3 270M7 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama1.0292L4 · 23 GB466—4 + 3 lang
IQ2_M by unsloth · ollama0.269Intel(R) Iris(R) Xe Graphics · 1 GB522—3 + 1 lang
IQ3_XXS by unsloth · ollama0.352Intel(R) Iris(R) Xe Graphics · 1 GB432—3 + 1 lang
IQ2_XXS by unsloth · ollama0.335Intel(R) Iris(R) Xe Graphics · 1 GB352—2
Q4_0 by unsloth · ollama0.333Linux, 4 CPUs · 31 GB—1—4 + 1 lang
Q8_0 by unsloth · ollama0.431Linux, 4 CPUs · 31 GB—132K3 + 1 lang
F16 by unsloth · ollama0.624Linux, 4 CPUs · 31 GB—116K3 + 1 lang

Gemma 3 1B6 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ggml-org · ollama1.014Linux, 4 CPUs · 31 GB1328K5 + 7 lang
Q3_K_S by maziyarpanahi · ollama0.814Linux, 4 CPUs · 31 GB—18K5 + 1 lang
Q3_K_L by lmstudio-community · ollama0.913Linux, 4 CPUs · 31 GB—18K5 + 5 lang
Q2_K by maziyarpanahi · ollama0.813Linux, 4 CPUs · 31 GB—18K5 + 1 lang
Q6_K by lmstudio-community · ollama1.211Linux, 4 CPUs · 31 GB—18K5 + 5 lang
Q8_0 by ggml-org · ollama1.29.7Linux, 4 CPUs · 31 GB—18K5 + 6 lang

gpt-oss

gpt-oss 20B6 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama1573146128K16 + 21 lang
Q3_K_S by unsloth · ollama15552× Tesla T4 · 15 GB4.4364K14 + 19 lang
Q8_K_XL by unsloth · ollama18442× Tesla T4 · 15 GB242128K16 + 10 lang
Q4_0 by unsloth · ollama124.6Linux, 4 CPUs · 31 GB4.234K14 + 17 lang
Q2_K by unsloth · ollama124.3Linux, 4 CPUs · 31 GB4.224K16 + 18 lang
Q8_0 by unsloth · ollama124.0Linux, 4 CPUs · 31 GB4.024K15 + 13 lang

OpenAI-gpt-oss-20B-Claude-4.5-Opus-Heretic-Uncensored3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_0 by mradermacher · ollama124.8Linux, 4 CPUs · 31 GB—1—0
Q3_K_S by mradermacher · ollama124.7Linux, 4 CPUs · 31 GB—1—0
IQ1_M by mradermacher · ollama124.2Linux, 4 CPUs · 31 GB—1—0

openai_gpt-oss-20b-GGUF-MXFP4-Experimental1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
MXFP4 by bartowski · ollama124.3Linux, 4 CPUs · 31 GB—1—0

gpt-oss-20b-heretic1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by DavidAU · ollama223.9—1—0

gpt-oss-20b-Derestricted1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q5_K_M by MaziyarPanahi · ollama173.4Linux, 4 CPUs · 31 GB—1—0

Huihui-gpt-oss-20b-BF16-abliterated1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by DavidAU · ollama223.3Linux, 4 CPUs · 31 GB—1—0

Qwen 3.5

Qwen 3.5 4B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama1242L4 · 23 GB232256K15 + 9 lang

Qwen 3.5 0.8B3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama0.8171624K11 + 1 lang
Q4_0 by unsloth · ollama0.817Linux, 4 CPUs · 31 GB—1—5 + 1 lang
Q8_0 by unsloth · ollama1.113Linux, 4 CPUs · 31 GB—1—4

Qwen 3.5 2B3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama—8.67.82—0
Q4_0 by unsloth · ollama—7.5—1—0
Q8_0 by unsloth · ollama—5.7—1—0

Qwen 3.5 35B-A3B4 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama224.2Linux, 4 CPUs · 31 GB3.924K10 + 15 lang
Q4_K_S by unsloth · ollama214.1Linux, 4 CPUs · 31 GB—14K9 + 9 lang
IQ2_M by unsloth · ollama—not timed—14K9 + 7 lang
IQ2_XXS by unsloth · ollama—not timed—1—9 + 7 lang

Qwen 3.5 9B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by unsloth · ollama—2.5—1—0
Q4_K_M by unsloth · ollama—2.4machine not reported2.32—0

MMS-TTS (Meta)

MMS-TTS Assamese1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.85.8s of audio in 5.8s · 1.0× slower than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Bengali1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.25.3s of audio in 5.4s · 1.0× slower than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Gujarati1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.46.1s of audio in 6.1s · 1.0× slower than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Hindi1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf8.85.5s of audio in 6.8s · 1.2× slower than realtime · on Windows, 8 cores, 16 GB—5—2

MMS-TTS Kannada1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.66.1s of audio in 5.9s · 1.0× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Malayalam1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.74.0s of audio in 4.5s · 1.1× slower than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Marathi1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.37.4s of audio in 7.1s · 1.0× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Odia1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.85.3s of audio in 5.2s · 1.0× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Punjabi1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.76.8s of audio in 6.4s · 1.1× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Tamil1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.39.1s of audio in 8.9s · 1.0× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Telugu1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.36.4s of audio in 6.1s · 1.1× faster than realtime · on Linux, 4 cores, 31 GB—1—2

MMS-TTS Urdu1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf1.89.0s of audio in 18s · 2.0× slower than realtime · on Linux, 4 cores, 31 GB—1—2

Qwen 3.6

Qwen 3.6 27B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama279.53.24128K16 + 9 lang

Qwen3.6-14B-A3B-FableVibes2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by tvall43 · ollama—5.6—1—0
Q3_K_M by tvall43 · ollama—0.8—1—0

Huihui-Qwen3.6-35B-A3B-abliterated-MTP-GGUF2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q2_K by huihui-ai · ollama135.0Linux, 4 CPUs · 31 GB—1—14 + 3 lang
Q6_K by huihui-ai · ollama294.3Linux, 4 CPUs · 31 GB4.124K14 + 19 lang

Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by llmfan46 · ollama224.9Linux, 4 CPUs · 31 GB—14K15 + 10 lang
Q4_K_S by llmfan46 · ollama204.9Linux, 4 CPUs · 31 GB—14K15 + 10 lang
Q3_K_M by llmfan46 · ollama174.8Linux, 4 CPUs · 31 GB—14K15 + 9 lang

Qwen 3.6 35B-A3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ1_M by unsloth · ollama—not timed—14K14 + 9 lang

Qwen 3.8

Huihui-Qwen3.8-27B-abliterated-GGUF2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_S by huihui-ai · ollama2513—1128K16 + 10 lang
Q4_K_M by huihui-ai · ollama2610.0—1128K16 + 10 lang

Dirk-Qwen3.8-27B-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by peculiar-ragdoll · ollama2610machine not reported—1128K21 + 10 lang

Qwen3.8-27B-Uncensored-GGUF2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by JonathanColetti · ollama2610machine not reported—1128K16 + 10 lang
Q5_K_M by JonathanColetti · ollama294.3—1128K16 + 10 lang

Qwen 3.8 27B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by unsloth · ollama322.92× Tesla T4 · 15 GB2.1232K16 + 16 lang
Q4_K_M by unsloth · ollama—not timed—28K11 + 10 lang

Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q6_K_P by HauhauCS · ollama311.8machine not reported—164K11 + 10 lang

Granite 4.0

Granite 4.0 SPEECH-1B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ibm-granite · ollama1.51302× Tesla T4 · 15 GB—1—2

Granite 4.0 1B6 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_0 by ibm-granite · ollama1.412—14K9 + 1 lang
Q4_K_M by ibm-granite · ollama1.411Linux, 4 CPUs · 31 GB1124K9 + 5 lang
Q3_K_S by ibm-granite · ollama1.28.5—14K7 + 1 lang
Q2_K by ibm-granite · ollama1.18.2Linux, 4 CPUs · 31 GB—1—2 + 1 lang
Q3_K_L by ibm-granite · ollama1.37.2Linux, 4 CPUs · 31 GB—14K9 + 1 lang
BF16 by ibm-granite · ollama3.74.7Linux, 4 CPUs · 31 GB—14K7 + 4 lang

GLM-OCR

GLM-OCR BASE5 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mradermacher · ollama2.2106L4 · 23 GB6024K7 + 1 lang
Q3_K_S by mradermacher · ollama2.118Intel(R) Iris(R) Xe Graphics · 1 GB—14K6 + 1 lang
Q2_K by mradermacher · ollama2.013Intel(R) Iris(R) Xe Graphics · 1 GB—1—6 + 1 lang
FP8 by axteam · vllm—not timed—14K10 + 1 lang
W4A16 by AlphaOxO · vllm—not timed—14K5 + 1 lang

SD-Turbo

base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP16 by Darylgeero · diffusers6.9512×512 in 49s · 0.02 steps/sec · on Linux, 4 cores, 31 GB—1—1

memorized-commons101 build

BuildGBFastest tok/sTypicalMachinesContextChecks
diffusers by nbkcnil · diffusers5.6512×512 in 37s · 0.03 steps/sec · on Linux, 4 cores, 31 GB—1—1

phi-lattice · ununununununununununununununununununun1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by ununununununununununununununununununun · generation10not timed—3—1

phi-lattice-v2 · ununununununununununununununununununun1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by ununununununununununununununununununun · generation9.0not timed—2—1

SD-Turbo (SD 2.1 distilled)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP16 by stabilityai · diffusers6.4512×512 in 53s · 0.02 steps/sec · on Linux, 4 cores, 31 GB—7—1

CogVideoX

5b1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
diffusers by zai-org · diffusers—not timed—1—0

CogVideoX 2B3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by THUDM · diffusers17not timed—4—0
build by harahan—not timed—1—0
diffusers by zai-org · diffusers—not timed—1—0

Hunyuan MT 2

Hunyuan MT 2 1.8B3 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by tencent · ollama2.8233254—0
Q3_K_S by mradermacher · ollama2.117—1—0
Q2_K by mradermacher · ollama—11—1—0

Hunyuan MT 2 30B-A3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by tencent · ollama195.4Linux, 4 CPUs · 31 GB4.524K7 + 10 lang

North 1.0

forge-1-mini · north-ml12 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q2_K by north-ml1 · ollama0.01075—1—0
Q3_K_M by north-ml1 · ollama0.0629—1—1

North 1.0 MINI-CODE-1-01 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama203.2Linux, 4 CPUs · 31 GB2.724K11 + 9 lang

North-Mini-Code-1.0-i1-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XXS by mradermacher · ollama122.3—14K11 + 9 lang

Whisper

Whisper large-v31 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP16 by openai · speech-hf—21s of audio in 15s · 1.4× faster than realtime · on Linux, 12 cores, 53 GB—1—2 + 10 lang

Whisper large-v3-turbo1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP16 by openai · speech-hf4.021s of audio in 3m 23s · 9.7× slower than realtime · on Linux, 4 cores, 31 GB—6—2 + 10 lang

Whisper medium1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by openai · speech-hf3.416s of audio in 3m 41s · 14.0× slower than realtime · on Windows, 8 cores, 16 GB—2—2 + 4 lang

Whisper small1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by openai · speech-hf2.619s of audio in 2m 29s · 7.9× slower than realtime · on Windows, 8 cores, 16 GB—3—2 + 6 lang

Whisper fine-tuned per Indian language (vasista22)

Whisper small hindi (vasista22)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by vasista22 · speech-hf2.019s of audio in 16m 51s · 53.6× slower than realtime · on Linux, 4 cores, 31 GB—1—2 + 1 lang

Whisper small kannada (vasista22)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by vasista22 · speech-hf5.06.4s of audio in 4m 40s · 43.6× slower than realtime · on Linux, 4 cores, 31 GB—1—2

Whisper small tamil (vasista22)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by vasista22 · speech-hf2.023s of audio in 19m 40s · 52.5× slower than realtime · on Linux, 4 cores, 31 GB—1—2 + 1 lang

Whisper small telugu (vasista22)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by vasista22 · speech-hf1.720s of audio in 37m 7s · 112.8× slower than realtime · on Linux, 4 cores, 31 GB—1—2 + 1 lang

Granite 4.2

Granite 4.2 3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ibm-granite · ollama1317L4 · 23 GB3.5332K16 + 3 lang

Granite 4.2 8B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ibm-granite · ollama—1.9machine not reported—1—0

Granite 4.2 30B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ibm-granite · ollama—0.7—1—0

Kimi Linear

Kimi Linear 48B-A3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bartowski · ollama304.8Linux, 4 CPUs · 31 GB3.924K14 + 9 lang

Kimi-Linear-REAP-35B-A3B-Instruct1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mradermacher · ollama214.7Linux, 4 CPUs · 31 GB—14K6 + 7 lang

Kimi Linear 48B-A3B-BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mradermacher · ollama—3.4—1—0

LFM2.5

LFM2.5 1.2B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama0.916Linux, 4 CPUs · 31 GB—14K12 + 1 lang

LFM2.5-2.6B-Qwen3.8-Turbo-Brilliance-Power-X12-NEO-MAX-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q8_0 by DavidAU · ollama—4.1—1—0

LFM2.5 24B-A2B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XXS by unsloth · ollama9.13.6Linux, 4 CPUs · 31 GB—14K16 + 9 lang

Qwen-Image

2-1-lightning · lightx2v1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by lightx2v · generation—not timed—1—0

pruna-2-1 · prunaai1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by prunaai · generation—not timed—1—0

tiny-random · optimum-intel-internal-testing1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by optimum-intel-internal-testing · generation—not timed—2—0

Sarvam

Sarvam-12 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bartowski · ollama5.41782288K4 + 9 lang
BF16 by sarvamai · vllm—not timed—1—3 + 5 lang

Sarvam-M1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bartowski · ollama1716Linux, 4 CPUs · 31 GB7.828K5 + 9 lang

Wan 2.1

1.3b · wan-ai2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
build by wan-ai—not timed—3—0
build by alibaba-pai—not timed—2—0

Wan 2.1 T2V 1.3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by Wan-AI · diffusers—not timed—1—1

Depth Anything

Depth Anything V2 Base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by depth-anything · vision-task7.8230×86 in 28s · on Windows, 8 cores, 16 GB—8—1

Depth Anything V2 Small1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by depth-anything · vision-task7.8230×86 in 1.1s · on Windows, 8 cores, 16 GB—8—1

ERNIE 4.5

ernie-4.5-moe-tiny-random1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by yujiepan · vllm—not timed—1—0 + 1 lang

ernie-4.5-tiny-random1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by yujiepan · vllm—not timed—1—0 + 1 lang

F5-TTS

F5-Hindi 24 kHz (SPRING Lab)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by SPRINGLab · speech-hf1118s of audio in 9m 7s · 31.0× slower than realtime · on Windows, 8 cores, 16 GB—5—2

F5-TTS v1 Base (zero-shot voice, en/zh)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by SWivid · speech-hf116.8s of audio in 22s · 3.2× slower than realtime · on Linux, 12 cores, 53 GB—1—2 + 1 lang

Hunyuan MT 1.5

HY-MT1.5-1.8B-i1-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_0 by mradermacher · ollama2.31321322—0

Hunyuan MT 1.5 1.8B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by tencent · ollama2.346L4 · 23 GB272—0

LTX-Video

13b · lightricks1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
build by lightricks—not timed—1—0

LTX-Video 2B 0.9.51 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by Lightricks · diffusers—not timed—1—0

MiniCPM-o 4.5

MiniCPM-o 4.5 BASE2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama9.720L4 · 23 GB—18K11 + 2 lang
Q5_0 by openbmb · ollama—1.0—1—0

MiniCPM-V 4.6

MiniCPM-V 4.6 BASE2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama5.0144273256K16 + 1 lang
Q2_K by mradermacher · ollama—17—1—0

MiniCPM-V 4.7

MiniCPM-V 4.7 1B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama14221L4 · 23 GB—1256K12 + 2 lang
Q8_0 by openbmb · ollama14186L4 · 23 GB1002256K12 + 2 lang

MiniCPM4.1

MiniCPM4.1 8B2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama7.746L4 · 23 GB—164K5 + 5 lang
IQ3_XXS by mradermacher · ollama—2.2—1—0

Ministral 3

Ministral 3 3B-INSTRUCT-25121 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mistralai · ollama3117L4 · 23 GB4.7364K10 + 2 lang

Ministral 3 14B-INSTRUCT-25121 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mistralai · ollama9.11.8—14K13

Moonshine

moonshine base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
speech-pod by UsefulSensors · speech-pod—16s of audio in 3.4s · 4.7× faster than realtime · on Linux, 4 cores, 31 GB—5—2 + 1 lang

moonshine tiny1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
speech-pod by UsefulSensors · speech-pod0.716s of audio in 1.7s · 9.4× faster than realtime · on Linux, 4 cores, 31 GB—6—2 + 1 lang

MusicGen

MusicGen Medium1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · audiogen15not timed—2—1

MusicGen Small1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · audiogen124.0s of audio in 1m 56s · 28.9× slower than realtime · on Windows, 8 cores, 16 GB—11—1

NLLB-200 (Meta)

NLLB-200 distilled 1.3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · translate-hf15not timed—3—13

NLLB-200 distilled 600M1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · translate-hf13not timed—8—13

Qwen2.5

Qwen2.5 0.5B Instruct2 builds

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by Qwen · ollama3.9248GeForce RTX 3080 Ti · 12 GB31932K10 + 6 lang
ollama by ollama · ollama—4.8—1—3

Sana

600m · beh12221 build

BuildGBFastest tok/sTypicalMachinesContextChecks
generation by beh1222 · generation14512×512 in 6.6s · 0.15 steps/sec · on Linux, 12 cores, 53 GB—1—1

Sana 1.6B 1024px1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP16 by Efficient-Large-Model · diffusers13not timed—4—1

Stable Audio Open

open-1-0-model · languaman1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
build by languaman—not timed—1—0

open-models · aemotionstudio1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
build by aemotionstudio—not timed—1—0

Airavata

Airavata 7B (Hindi)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mradermacher · ollama13531934K3 + 2 lang

DeepSeek V3.2

Devstral-Small-2505-Deepseek-V3.2-Speciale-Distill1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XS by TeichAI · ollama110.4Linux, 4 CPUs · 31 GB—14K6 + 6 lang

DeepSeek-OCR

DeepSeek-OCR BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by sabafallah · ollama4.7340L4 · 23 GB1772—3 + 1 lang

DeepSeek-OCR 2

DeepSeek-OCR 2 BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by sabafallah · ollama5.2412GeForce RTX 3080 Ti · 12 GB3353—4 + 1 lang

ECAPA-TDNN (SpeechBrain)

spkrec-ecapa-voxceleb1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by speechbrain · speaker-id3.4not timed—3—1

EmbeddingGemma 2

EmbeddingGemma 2 BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by ngquocvinh · ollama0.4not timed—1—1

GLM-4.5

GLM-4.5-Air-0.72B-MTP1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by nm-testing · vllm—not timed—1—0

GLM-4.7

GLM-4.7 FLASH1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama282132.214128K16 + 16 lang

GLM-5

GLM-5_4layer1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by mradermacher · ollama—7.7—1—0

Granite 4.1

Granite 4.1 3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama2418L4 · 23 GB—164K10 + 3 lang

Hunyuan Hy3

hy3-tiny-random1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by yujiepan · vllm—not timed—1—0

HunyuanVideo

HunyuanVideo T2V 720p1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by hunyuanvideo-community · diffusers—not timed—1—0

Kimi K2.5

Qwen3.5-27B-kimi-k2.5-Reasoning-Distilled-i1-GGUF1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ3_XXS by mradermacher · ollama—0.4—1—0

Kokoro

kokoro base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
speech-pod by kokoro · speech-pod1.94.2s of audio in 2.4s · 1.7× faster than realtime · on Windows, 8 cores, 16 GB—6—2 + 1 lang

Krutrim

Krutrim-2 Instruct1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bartowski · ollama2015L4 · 23 GB14364K6 + 10 lang

MADLAD-400 MT (Google)

MADLAD-400 3B MT1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by google · translate-hf14not timed—2—11

Magistral Small

Magistral Small 25061 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama215.5L4 · 23 GB—140K11 + 10 lang

MedGemma 1.5

MedGemma 1.5 4B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by unsloth · ollama3.675L4 · 23 GB—164K9 + 9 lang

MiniCPM-V 4

MiniCPM-V 4 BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama3.097L4 · 23 GB3.73—10 + 1 lang

MiniCPM-V 4.5

MiniCPM-V 4.5 BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama1511L4 · 23 GB—116K18 + 1 lang

MiniCPM5

MiniCPM5 1B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by openbmb · ollama7.4381GeForce RTX 3080 Ti · 12 GB115464K15 + 1 lang

MiniMax-M2.1

MiniMax-M2.1 REAP-50-I11 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ1_S by mradermacher · ollama—1.0—1—0

MiniMax-M2.7

MiniMax-M2.7-abliterated-BF161 build

BuildGBFastest tok/sTypicalMachinesContextChecks
IQ1_S by mradermacher · ollama—1.4—1—0

Muse Glimmer

Muse Glimmer 30B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by meta-models · ollama156.9L4 · 23 GB1.33—0

OpenHathi

OpenHathi 7B Hindi base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by QuantFactory · ollama1352302—1

Qianfan-OCR

Qianfan-OCR BASE1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by Reza2kn · ollama2318L4 · 23 GB—132K14 + 6 lang

Sarvam-Translate

Sarvam-Translate (Gemma 3 4B)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
BF16 by sarvamai · translate-hf20not timed—2—14

Seamless M4T v2 (Meta)

Seamless M4T v2 Large (speech → text)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · speech-hf4.812s of audio in 2m 8s · 11.1× slower than realtime · on Windows, 8 cores, 16 GB—3—2 + 13 lang

Segment Anything

SAM ViT-Base1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by facebook · vision-task1.85 frames at 230×86 in 12m 7s · on Windows, 8 cores, 16 GB—9—1

SmolLM3

SmolLM3 3B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bartowski · ollama7.378L4 · 23 GB—164K6 + 2 lang

Tiny SD

Tiny SD1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by segmind · diffusers4.8512×512 in 74s · 0.02 steps/sec · on Windows, 8 cores, 16 GB—8—1

TranslateGemma

TranslateGemma 4B1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
Q4_K_M by bullerwins · ollama2.978L4 · 23 GB412128K5 + 7 lang

WeSpeaker

wespeaker-voxceleb-resnet34-LM1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by pyannote · speaker-id3.4not timed—3—1

XTTS

XTTS-v2 (17 languages, 6-second clone)1 build

BuildGBFastest tok/sTypicalMachinesContextChecks
FP32 by coqui · speech-hf4.26.2s of audio in 28s · 4.5× slower than realtime · on Linux, 12 cores, 53 GB—1—2