Models

Manifests for models a gezel crew can run locally: verified file hashes, per-engine download blocks (llama.cpp, MLX, Ollama), honest size and context numbers, license classes, and tuned sampling defaults. Sorted by how strongly the catalog recommends them.

Chat models

Model	Params	Quant	Context	License	Size	Engines	Reco
Qwen 3.6 (27B, Q4)	27B	Q4_K_M	262K	Apache 2.0	17.1 GB	llama.cpp, MLX, Ollama	20
Qwen 3.6 (27B, Q8)	27B	Q8_0	262K	Apache 2.0	29.0 GB	llama.cpp, MLX	20
Qwen 3.6 (35B-A3B, Q4)	35B	UD-Q4_K_M	262K	Apache 2.0	22.7 GB	llama.cpp, MLX	20
Qwen 3.6 (35B-A3B, Q8)	35B	Q8_0	262K	Apache 2.0	37.8 GB	llama.cpp, MLX	20
Qwen 3.8 (27B, IQ1_S)	27B	UD-IQ1_S	262K	Apache 2.0	6.2 GB	llama.cpp	20
Qwen 3.8 (27B, Q2)	27B	UD-Q2_K_XL	262K	Apache 2.0	9.8 GB	llama.cpp	20
Qwen 3.8 (27B, Q3)	27B	UD-Q3_K_XL	262K	Apache 2.0	13.1 GB	llama.cpp	20
Qwen 3.8 (27B, Q4)	27B	UD-Q4_K_M	262K	Apache 2.0	16.5 GB	llama.cpp, MLX	20
Qwen 3.8 (27B, Q6)	27B	UD-Q6_K	262K	Apache 2.0	22.0 GB	llama.cpp, MLX	20
Qwen 3.8 (27B, Q8)	27B	Q8_0	262K	Apache 2.0	29.0 GB	llama.cpp, MLX	20
Gemma 4 (12B, Q4)	12B	UD-Q4_K_XL	256K	Apache 2.0	6.7 GB	llama.cpp, MLX	15
Gemma 4 (12B, Q8)	12B	Q8_0	256K	Apache 2.0	12.7 GB	llama.cpp, MLX	15
Gemma 4 (26B, MoE, Q4)	25.2B	UD-Q4_K_XL	256K	Apache 2.0	14.2 GB	llama.cpp, MLX, Ollama	15
Gemma 4 (31B, Q4)	30.7B	UD-Q4_K_XL	256K	Apache 2.0	17.3 GB	llama.cpp, MLX, Ollama	15
Gemma 4 (E2B, Q4)	2.3B	UD-Q4_K_XL	128K	Apache 2.0	2.6 GB	llama.cpp, MLX, Ollama	15
Gemma 4 (E4B, Q4)	8B	UD-Q4_K_XL	128K	Apache 2.0	4.2 GB	llama.cpp, MLX, Ollama	15
BTL-4 (35B-A3B, Q4)	35B	Q4_K_M	262K	Apache 2.0	21.4 GB	llama.cpp	10
BTL-4 Compact (35B-A3B, IQ2_XXS)	35B	IQ2_XXS	262K	Apache 2.0	10.0 GB	llama.cpp	10
DeepSeek R1 (8B, Q4)	8B	Q4_K_M	128K	MIT	5.2 GB	llama.cpp, MLX, Ollama
DeepSeek V4 Flash (FP4)	284B	FP4-FP8	1000K	MIT	164.6 GB	llama.cpp, MLX
DeepSeek V4 Flash (IQ2_XXS)	284B	IQ2_XXS	1000K	MIT	86.7 GB	llama.cpp
DeepSeek V4 Flash (mixed 2/4-bit)	284B	IQ2_XXS/Q4_K	1000K	MIT	97.6 GB	llama.cpp
DeepSeek V4 Flash (MXFP4)	284B	MXFP4	1000K	MIT	156.0 GB	llama.cpp
GLM 5.2 (IQ2_XXS)	754B		1049K	MIT	211.1 GB
GLM 5.3 Flash (320B-A18B, IQ2_XXS)	320B		1049K	MIT	97.6 GB
GPT-OSS (120B, MoE, Q4)	117B	Q4_K_M	131K	Apache 2.0	65.0 GB	llama.cpp, MLX, Ollama
GPT-OSS (20B, MoE, Q4)	20B	Q4_K_M	131K	Apache 2.0	12.0 GB	llama.cpp, MLX, Ollama
Granite 4.2 (30B, Q4)	30B	Q4_K_M	131K	Apache 2.0	17.7 GB	llama.cpp, MLX
Granite 4.2 (3B, Q4)	3B	Q4_K_M	131K	Apache 2.0	2.2 GB	llama.cpp, MLX
Granite 4.2 (8B, Q4)	8B	Q4_K_M	131K	Apache 2.0	5.3 GB	llama.cpp, MLX
Laguna S 2.1 (118B-A8B, Q4)	118B	Q4_K_M	262K	OpenMDW 1.1	96.0 GB	llama.cpp, MLX
Laguna S 2.1 (118B-A8B, Q6 MLX)	118B	oQ6e	262K	OpenMDW 1.1	92.0 GB	MLX
Laguna S 2.1 (118B-A8B, Q8)	118B	Q8_0	262K	OpenMDW 1.1	128.8 GB	llama.cpp, MLX
LFM2.5 (2.6B, Q4)	2.6B	Q4_K_M	128K	LFM Open 1.0	1.7 GB	llama.cpp, MLX
Llama 3.2 (3B, Q4)	3B	Q4_K_M	128K	Llama 3.2 Community	2.0 GB	llama.cpp, MLX, Ollama
Mistral 7B Instruct (Q4)	7B	Q4_K_M	33K	Apache 2.0	4.4 GB	llama.cpp, MLX, Ollama
Mistral Medium 3.5 (128B, Q4)	128B	Q4_K_M	262K	MIT-Modified	74.9 GB	llama.cpp, MLX
Muse Glimmer (30B, Q4)	30B	Q4_K_M	131K	Apache 2.0	16.8 GB	llama.cpp
Nanbeige 4.2 (3B, Q4)	3B	Q4_K_M	262K	Apache 2.0	2.7 GB	llama.cpp
Nemotron 3.5 Lightning (30B-A3B, Q4)	30B	Q4_K_M	262K	OpenMDW 1.1	25.4 GB	llama.cpp, MLX
Nemotron 3.5 Lightning (30B-A3B, Q8)	30B	Q8_0	262K	OpenMDW 1.1	35.0 GB	llama.cpp, MLX
Nemotron-3 Nano (30B, MoE Reasoning, Q4)	30B	Q4_K_M	131K	NVIDIA Open Model	24.5 GB	llama.cpp, MLX
Nemotron-3 Super (120B-A12B, Q4)	120B	Q4_K_M	131K	NVIDIA Open Model	86.1 GB	llama.cpp
Ornith 1.0 (35B, Q4)	35B	Q4_K_M	262K	MIT	21.2 GB	llama.cpp, MLX
Ornith 1.0 (35B, Q8)	35B	Q8_0	262K	MIT	37.0 GB	llama.cpp, MLX
Ornith 1.0 (9B, Q4)	9B	Q4_K_M	262K	MIT	5.6 GB	llama.cpp, MLX
Ornith 1.5 (35B-A3B, Q4)	35B	Q4_K_M	262K	MIT	21.7 GB	llama.cpp, MLX
Ornith 1.5 (35B-A3B, Q8)	35B	Q8_0	262K	MIT	36.9 GB	llama.cpp, MLX
Ornith 1.5 (9B, Q4)	9B	Q4_K_M	262K	MIT	5.6 GB	llama.cpp, MLX
Ornith 1.5 (9B, Q8)	9B	Q8_0	262K	MIT	9.5 GB	llama.cpp, MLX
Qwen 3.5 (122B-A10B, Q4)	122B	UD-Q4_K_M	262K	Apache 2.0	78.3 GB	llama.cpp, MLX
Qwen 3.5 (27B, Q4)	27B	Q4_K_M	262K	Apache 2.0	17.1 GB	llama.cpp, MLX, Ollama
Qwen 3.5 (2B, Q4)	2B	Q4_K_M	256K	Apache 2.0	1.3 GB	llama.cpp, MLX, Ollama
Qwen 3.5 (4B, Q4)	4B	Q4_K_M	256K	Apache 2.0	2.8 GB	llama.cpp, MLX, Ollama
Qwen 3.5 (9B, Q4)	9B	Q4_K_M	262K	Apache 2.0	5.9 GB	llama.cpp, MLX, Ollama
Qwen 3.8 Flash Next (180B-A6B, IQ3_XXS)	180B	UD-IQ3_XXS	262K	Qwen Community 1.0	82.0 GB	llama.cpp
Qwen 3.8 Flash Next (180B-A6B, IQ4_XS)	180B	UD-IQ4_XS	262K	Qwen Community 1.0	93.7 GB	llama.cpp
Talkie 1930 (13B, Q4)	13B	Q4_K_M	2K	Apache 2.0	8.6 GB	llama.cpp
Ternary Bonsai 27B (2-bit)	27B	2bit	262K	Apache 2.0	8.5 GB	MLX

Image models

Model	Category	Quant	License	Size	Hardware
Animagine XL 4.0	artistic	FP16	Fair AI 1.0-SD	6.9 GB	mid
FLUX.1 dev (Q4)	general	Q4_K_S	FLUX.1 dev	6.8 GB	high
FLUX.1 schnell (Q4)	general	Q4_K_S	Apache 2.0	6.8 GB	high
FLUX.2 Klein 4B (Q4)	general	Q4_0	Apache 2.0	2.5 GB	high
Krea 2 Turbo (Q4)	general	Q4_K_M	Krea 2 Community	7.5 GB	high
Krea 2 Turbo (Q6)	general	Q6_K	Krea 2 Community	10.6 GB	workstation
Krea 2 Turbo (Q8)	general	Q8_0	Krea 2 Community	13.7 GB	workstation
SD-Turbo	general	FP16	Stability Community	5.2 GB	low
SDXL Lightning (4-step)	general	FP16	OpenRAIL++	6.9 GB	mid
SDXL Turbo	general	FP16	Stability NC Research	6.9 GB	mid
Stable Diffusion 1.5	general	FP16	OpenRAIL-M	4.3 GB	low
Stable Diffusion XL 1.0	general	FP16	OpenRAIL++-M	6.9 GB	mid

Video models

Model	Family	Frames	License	Size	Hardware
LTX-2 (19B, video + audio)	ltx2	121	LTX Open Weights	92.1 GB	high
LTX-Video 0.9.7 dev (13B)	ltx	97	LTX Open Weights	47.6 GB	high
LTX-Video 0.9.8 distilled (13B)	ltx	97	LTX Open Weights	47.6 GB	high
Wan 2.2 TI2V-5B	wan	121	Apache 2.0	34.2 GB	high