Open models
Publicly released model artifacts, weights, and checkpoints; open does not imply an OSI-approved license. Classify by primary task: embeddings or reranking first, then specialized speech/audio or image/video tasks, then general multimodal models, then text/code language models. Genuinely mixed releases may use this parent. Excludes robot-specific embodied models, reusable execution engines, training tools, and authoring applications.
Subfields
Language models
Released text- or code-focused language-model weights and checkpoints. Excludes task-specialized embedding or reranking models, speech/audio models, image/video models, general multimodal models, training frameworks, execution engines, and coding-agent applications.
3 projectsMultimodal models
Released general-purpose vision-language or omni-model weights combining text with image, video, or audio understanding or interaction. Specialized embedding, reranking, speech/audio, and image/video tasks take precedence. Excludes robot-specific embodied models, execution engines, and user-facing multimodal workspaces.
1 projects| Project | What it does | Stars | Language | License | Latest release |
|---|---|---|---|---|---|
| nandhakishorm/laya | A multilingual non-autoregressive model that returns choices, scores, and yes/no decisions in one forward pass. | ★ 33.3K | Python | Apache-2.0 | v0.4.11 |
| startluxlabs/startlux-decision | Releases public 0.8B–27B decision-model weights in GGUF and MLX formats for calibrated option probabilities. | ★ 676 | Python | Apache-2.0 | v1.1.0 |
| rockmatrix-ai/edgerock | Provides a public multimodal model for long reasoning with constant-size persistent history. | ★ 488 | Python | Apache-2.0 | — |
| liqiiiii/jeva | The released Jeva-4B model computes candidate probabilities, truth probabilities, and expected scores. | ★ 101 | Python | Apache-2.0 | — |