Every model
These are all the AI models in Gezel’s catalog. Each one runs on your own computer through an engine Gezel installs for you, and Gezel recommends the ones that fit your hardware when you set it up. The model scorecard shows how the models we test actually do on real jobs, and Local models and tiers explains how to choose.
Chat models (66)
The models your gezels think and write with, recommended first.
| Model | Size | Context | Tools | Best for | Runs on | License |
|---|---|---|---|---|---|---|
| Qwen 3.8 (27B, Q4) | 16.5 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.8 (27B, Q3) | 13.1 GB | 262K | Yes | Everyday work | llama.cpp | Apache 2.0 |
| Qwen 3.8 (27B, Q2) | 9.8 GB | 262K | Yes | Everyday work | llama.cpp | Apache 2.0 |
| Qwen 3.6 (27B, Q4) | 17.1 GB | 262K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Qwen 3.6 (27B, Q8) | 29.0 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.6 (35B-A3B, Q4) | 22.7 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.6 (35B-A3B, Q8) | 37.8 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.8 (27B, IQ1_S) | 6.2 GB | 262K | Yes | Everyday work | llama.cpp | Apache 2.0 |
| Qwen 3.8 (27B, Q6) | 22.0 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.8 (27B, Q8) | 29.0 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Gemma 4 (12B, Q4) | 6.7 GB | 256K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Gemma 4 (12B, Q8) | 12.7 GB | 256K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Gemma 4 (26B, MoE, Q4) | 14.2 GB | 256K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Gemma 4 (31B, Q4) | 17.3 GB | 256K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Gemma 4 (E2B, Q4) | 2.6 GB | 128K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Gemma 4 (E4B, Q4) | 4.2 GB | 128K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| BTL-4 (35B-A3B, Q4) | 21.4 GB | 262K | Yes | Coding | llama.cpp | Apache 2.0 |
| BTL-4 Compact (35B-A3B, IQ2_XXS) | 10.0 GB | 262K | Yes | Coding | llama.cpp | Apache 2.0 |
| DeepSeek R1 (8B, Q4) | 5.2 GB | 128K | Yes | Reasoning | llama.cpp, MLX, Ollama | MIT |
| DeepSeek V4 Flash (FP4) | 164.6 GB | 1000K | Yes | Reasoning | llama.cpp, MLX, DwarfStar | MIT |
| DeepSeek V4 Flash (IQ2_XXS) | 86.7 GB | 1000K | Yes | Reasoning | llama.cpp, DwarfStar | MIT |
| DeepSeek V4 Flash (mixed 2/4-bit) | 97.6 GB | 1000K | Yes | Reasoning | llama.cpp, DwarfStar | MIT |
| DeepSeek V4 Flash (MXFP4) | 156.0 GB | 1000K | Yes | Reasoning | llama.cpp, DwarfStar | MIT |
| GLM 5.2 (IQ2_XXS) | 211.1 GB | 1049K | Yes | Reasoning | DwarfStar | MIT |
| GLM 5.3 Flash (320B-A18B, IQ2_XXS) | 97.6 GB | 1049K | Yes | Reasoning | DwarfStar | MIT |
| GPT-OSS (120B, MoE, Q4) | 65.0 GB | 131K | Yes | Reasoning | llama.cpp, MLX, Ollama | Apache 2.0 |
| GPT-OSS (20B, MoE, Q4) | 12.0 GB | 131K | Yes | Reasoning | llama.cpp, MLX, Ollama | Apache 2.0 |
| Granite 4.2 (30B, Q4) | 17.7 GB | 131K | Yes | Reasoning | llama.cpp, MLX | Apache 2.0 |
| Granite 4.2 (3B, Q4) | 2.2 GB | 131K | Yes | Reasoning | llama.cpp, MLX | Apache 2.0 |
| Granite 4.2 (8B, Q4) | 5.3 GB | 131K | Yes | Reasoning | llama.cpp, MLX | Apache 2.0 |
| Laguna S 2.1 (118B-A8B, Q4) | 96.0 GB | 262K | Yes | Coding | llama.cpp, MLX | OpenMDW 1.1 |
| Laguna S 2.1 (118B-A8B, Q6 MLX) | 92.0 GB | 262K | Yes | Coding | MLX | OpenMDW 1.1 |
| Laguna S 2.1 (118B-A8B, Q8) | 128.8 GB | 262K | Yes | Coding | llama.cpp, MLX | OpenMDW 1.1 |
| LFM2.5 (2.6B, Q4) | 1.7 GB | 128K | Yes | Everyday work | llama.cpp, MLX | LFM Open 1.0 |
| LFM2.5 (24B-A2B, Q4) | 13.6 GB | 128K | Yes | Everyday work | llama.cpp | LFM Open 1.0 |
| Llama 3.2 (3B, Q4) | 2.0 GB | 128K | Yes | Everyday work | llama.cpp, MLX, Ollama | Llama 3.2 Community |
| MiniCPM-V 4.7 (35B-A3B, Q4) | 22.3 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| MiniCPM5 (1B, Q4) | 0.7 GB | 131K | Yes | Everyday work | llama.cpp, MLX | Apache-2.0 |
| MiniCPM5 (2B, Q4) | 1.6 GB | 131K | Yes | Everyday work | llama.cpp, MLX | Apache-2.0 |
| Mistral 7B Instruct (Q4) | 4.4 GB | 33K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Mistral Medium 3.5 (128B, Q4) | 74.9 GB | 262K | Yes | Reasoning | llama.cpp, MLX | MIT-Modified |
| Muse Glimmer (30B, Q4) | 16.8 GB | 131K | Yes | Everyday work | llama.cpp | Apache 2.0 |
| Nanbeige 4.2 (3B, Q4) | 2.7 GB | 262K | Yes | Reasoning | llama.cpp | Apache 2.0 |
| Nemotron 3.5 Lightning (30B-A3B, Q4) | 25.4 GB | 262K | Yes | Reasoning | llama.cpp, MLX | OpenMDW 1.1 |
| Nemotron 3.5 Lightning (30B-A3B, Q8) | 35.0 GB | 262K | Yes | Reasoning | llama.cpp, MLX | OpenMDW 1.1 |
| Nemotron-3 Nano (30B, MoE Reasoning, Q4) | 24.5 GB | 131K | Yes | Reasoning | llama.cpp, MLX | NVIDIA Open Model |
| Nemotron-3 Super (120B-A12B, Q4) | 86.1 GB | 131K | Yes | Reasoning | llama.cpp | NVIDIA Open Model |
| Ornith 1.0 (35B, Q4) | 21.2 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.0 (35B, Q8) | 37.0 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.0 (9B, Q4) | 5.6 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.5 (35B-A3B, Q4) | 21.7 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.5 (35B-A3B, Q8) | 36.9 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.5 (9B, Q4) | 5.6 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Ornith 1.5 (9B, Q8) | 9.5 GB | 262K | Yes | Coding | llama.cpp, MLX | MIT |
| Qwen 3.5 (0.8B, Q4) | 0.5 GB | 256K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache-2.0 |
| Qwen 3.5 (122B-A10B, Q4) | 78.3 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Apache 2.0 |
| Qwen 3.5 (27B, Q4) | 17.1 GB | 262K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Qwen 3.5 (2B, Q4) | 1.3 GB | 256K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Qwen 3.5 (4B, Q4) | 2.8 GB | 256K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Qwen 3.5 (9B, Q4) | 5.9 GB | 262K | Yes | Everyday work | llama.cpp, MLX, Ollama | Apache 2.0 |
| Qwen 3.8 Flash Next (180B-A6B, 4-bit) | 93.7 GB | 262K | Yes | Everyday work | llama.cpp, MLX | Qwen Community 1.0 |
| Qwen 3.8 Flash Next (180B-A6B, IQ2_XXS/Q2_K, DS4) | 147.8 GB | 262K | Yes | Everyday work | DwarfStar | Qwen Community 1.0 |
| Qwen 3.8 Flash Next (180B-A6B, IQ3_XXS) | 82.0 GB | 262K | Yes | Everyday work | llama.cpp | Qwen Community 1.0 |
| Talkie 1930 (13B, Q4) | 8.6 GB | 2K | No | Everyday work | llama.cpp | Apache 2.0 |
| Ternary Bonsai 27B (2-bit) | 8.5 GB | 262K | Yes | Everyday work | MLX | Apache 2.0 |
| Underdog Saluki 1.0 (27B, IQ2 mix) | 7.9 GB | 262K | Yes | Everyday work | llama.cpp | Apache 2.0 |
Image models (12)
Models your crew can use to draw pictures, icons and illustrations. Gezel works out what each one needs from its files and offers only the ones your machine can run.
| Model | Size | Commercial use | License |
|---|---|---|---|
| Animagine XL 4.0 | 6.9 GB | Yes | Fair AI 1.0-SD |
| FLUX.2 Klein 4B (Q4) | 2.5 GB | Yes | Apache 2.0 |
| Iris 3B (F32) | 12.0 GB | Yes | Apache 2.0 |
| Krea 2 Turbo (Q4) | 7.5 GB | No | Krea 2 Community |
| Krea 2 Turbo (Q6) | 10.6 GB | No | Krea 2 Community |
| Krea 2 Turbo (Q8) | 13.7 GB | No | Krea 2 Community |
| Ming Image 0.1 Design (BF16) | 12.3 GB | Yes | MIT |
| SD-Turbo | 5.2 GB | Yes | Stability Community |
| SDXL Lightning (4-step) | 6.9 GB | Yes | OpenRAIL++ |
| SDXL Turbo | 6.9 GB | No | Stability NC Research |
| Stable Diffusion 1.5 | 4.3 GB | Yes | OpenRAIL-M |
| Stable Diffusion XL 1.0 | 6.9 GB | Yes | OpenRAIL++-M |
Video models (6)
Models that turn a prompt or a picture into a short clip. Gezel works out what each one needs from its files and offers only the ones your machine can run.
| Model | Size | Longest clip | Commercial use | License |
|---|---|---|---|---|
| LTX-2 (19B, video + audio) | 92.1 GB | 11 s | Yes | LTX-2 Community |
| LTX-2.3 (22B, video + audio) | 94.9 GB | 11 s | Yes | LTX-2 Community |
| LTX-2.3 distilled (22B, video + audio) | 94.9 GB | 11 s | Yes | LTX-2 Community |
| LTX-Video 0.9.7 dev (13B) | 47.6 GB | 11 s | Yes | LTX Open Weights |
| LTX-Video 0.9.8 distilled (13B) | 47.6 GB | 11 s | Yes | LTX Open Weights |
| Wan 2.2 TI2V-5B | 34.2 GB | 5 s | Yes | Apache 2.0 |