image
qwen-image-2.1:7b
Qwen Image 2.1 7B: generation, transparent images and multi-reference editing
Arena Text-to-Image: #19 (1222.93)
Arena Image Edit: #18 (1368.69)
Appearance
Model registry
The registry is the source of truth for downloadable models. Search the catalog or download models.yml.
30 modelsDownload models.yml
image
qwen-image-2.1-turbo:7b
Qwen Image 2.1 Turbo 7B: eight-step generation and multi-reference image editing
image
flux2-klein:9b
FLUX.2-klein 9B: four-step generation and multi-reference image editing
Arena Text-to-Image: #62 (1070.29)
Arena Image Edit: #42 (1224.74)
image
flux3-klein:5.6b
FLUX.3 Image Klein 5.6B: sixteen-step generation and multi-reference image editing
decision
winnow:12b
Winnow 12B: Gemma 4 typed decisions about text and images
Decision Index: #47 (49.27)
JevBench: #11 (68.89)
decision
winnow:4b
Winnow E4B: efficient Gemma 4 typed decisions about text and images
Decision Index: #87 (38.73)
text
qwen3.5:9b
Qwen3.5 9B: multilingual chat model with reasoning and image input
text
gemma4:4b
Gemma 4 E4B: efficient multilingual chat model with reasoning and image input
text
gemma4:12b
Gemma 4 12B: multilingual chat model with reasoning and image input
embedding
clm-v0.1:8b
CLM v0.1 8B: contrastive encoder for text ranking and embeddings
embedding
nomic-embed-text-v1.5:137m
Nomic Embed v1.5: English text embeddings
MTEB: Listed; rank not imported
embedding
nomic-embed-text-v2-moe:475m
Nomic Embed v2 MoE: multilingual mixture-of-experts embeddings
MTEB: Listed; rank not imported
embedding
embeddinggemma-2:740m
EmbeddingGemma 2 740M: multilingual text, image and audio embeddings
audio
qwen3-tts:1.7b
Qwen3-TTS 1.7B Base: multilingual text-to-speech with reference audio
audio
pocket-tts-en:100m
Pocket TTS English: small CPU-friendly text-to-speech with voice cloning
audio
pocket-tts-de:100m
Pocket TTS German: small CPU-friendly text-to-speech with voice cloning
audio
pocket-tts-es:100m
Pocket TTS Spanish: small CPU-friendly text-to-speech with voice cloning
audio
pocket-tts-fr:100m
Pocket TTS French (24 layers): text-to-speech with voice cloning
audio
pocket-tts-it:100m
Pocket TTS Italian: small CPU-friendly text-to-speech with voice cloning
audio
pocket-tts-pt:100m
Pocket TTS Portuguese: small CPU-friendly text-to-speech with voice cloning
decision
clef-flash:9b
Clef Flash 9B: multimodal typed decisions with a joint schema head
Decision Index: #54 (47.61)
JevBench: #44 (36.64)
decision
kev:0.5b
Kev 0.5b: Qwen2.5-0.5B-based System One decision model (ggmlc)
decision
kev:0.8b
Kev 0.8b: Qwen3.5-0.8B-based System One decision model (ggmlc)
Decision Index: #133 (13.66)
decision
kev:4b
Kev 4b: Qwen3.5-4B-based System One decision model (ggmlc)
Decision Index: #84 (39.50)
decision
laya-english:421m
Laya English: ModernBERT encoder, English (ggmlc)
Decision Index: #150 (4.43)
JevBench: #127 (0.01)
decision
laya-multilingual:322m
Laya Multilingual: mmBERT encoder, multilingual (ggmlc)
decision
laya-typed-decisions:421m
Laya Typed Decisions: encoder tuned for typed decisions (ggmlc)
decision
decider:0.8b
Decider 0.8B: small Qwen3.5-based one-pass decision model
decision
decider:4b
Decider 4B: Qwen3.5-based one-pass decision model (v2.1)
Decision Index: #80 (40.33)
decision
decider-vision:2b
Decider 2B Vision: one-pass decisions about text and one image
Quality benchmarks
| Model type | Primary benchmark |
|---|---|
| text | MMLU-Pro |
| decision | Decision Index |
| embedding | MTEB |
| image | Arena Text-to-Image |
| audio | Seed-TTS test set |
Benchmark selection and verified model coverage.
Ranks require an exact verified upstream model identity, not a GGUF repository. MTEB membership is not a suite rank; HF model-card scores are publisher-reported and may use different datasets. Only confirmed benchmark entries are displayed. These results do not evaluate local quantizations.