Language models
Llama 3.1 / 3.2 / 3.3
General-purpose instruct models from 1B to 70B parameters
Llama Community License; acceptable-use terms apply
Models
Models we fine-tune
We fine-tune open-weight models on your data. The right base model depends on license terms, language coverage, latency targets, and where you plan to deploy.
License summaries are for orientation only and can change between releases. Confirm terms with your counsel before training on production data.
Language models
Mistral / Mixtral / Ministral
Dense and mixture-of-experts models for efficient instruction tuning
Apache 2.0 for most releases; some variants use Mistral licenses
Language models
Qwen2.5 / Qwen3
Multilingual, coding, and long-context workloads
Apache 2.0 for most sizes; some Qwen2.5 sizes use the Qwen license
Language models
Gemma 2 / Gemma 3
Compact models for cost-sensitive and on-device serving
Gemma Terms of Use
Language models
Phi-3 / Phi-4
Small models for edge deployment and tight latency budgets
MIT
Language models
DeepSeek-R1 distills
Reasoning-oriented models from 1.5B to 70B parameters
MIT; distills also follow their Qwen or Llama base terms
Language models
gpt-oss-20b / 120b
Open-weight reasoning models for private deployment
Apache 2.0 with a usage policy
Language models
OLMo 2
Open data, code, and training recipes for reproducible work
Apache 2.0
Language models
Falcon 3
Efficient 1B to 10B models for modest hardware
TII Falcon License
Language models
SmolLM2 / SmolLM3
Compact models for on-device and low-cost serving
Apache 2.0
Language models
Mistral Small 3 / Mistral NeMo
Mid-size instruct models with strong quality per GPU
Apache 2.0
Language models
IBM Granite 3.x
Enterprise-oriented models for RAG, tool use, and summarization
Apache 2.0
Language models
Qwen2.5-Coder
Code models from 0.5B to 32B for completion and review
Apache 2.0 for most sizes; the 3B model uses a research license
Language models
Yi-1.5
Bilingual English and Chinese assistants
Apache 2.0
Language models
InternLM 2.5
Bilingual models with strong long-context behavior
InternLM license; confirm commercial terms
Language models
StarCoder2 / Code Llama
Code completion and review on private repositories
BigCode OpenRAIL-M; Llama 2 Community License
Language models
DeepSeek-Coder
Code models for completion and refactoring
DeepSeek Model License; commercial use permitted
Language models
Nous Hermes 3
Instruction-tuned Llama 3.1 variants for further domain tuning
Llama 3.1 Community License
Vision-language
Llama 3.2 Vision (11B / 90B)
Image and text question answering, document understanding
Llama Community License; multimodal use restricted for EU-domiciled entities
Vision-language
Qwen2-VL / Qwen2.5-VL
Documents, charts, diagrams, and product imagery
Apache 2.0 for most sizes; Qwen license on the largest
Vision-language
Pixtral 12B
Document, chart, and natural-image understanding
Apache 2.0
Vision-language
Molmo
Open vision-language models with pointing and counting abilities
Apache 2.0
Vision-language
SmolVLM
Small vision-language models for low-cost and on-device use
Apache 2.0
Vision-language
Phi-3.5 Vision / Phi-4 Multimodal
Compact image, document, and audio-aware models
MIT
Vision-language
InternVL2
High-resolution document and chart understanding
MIT; base model terms also apply
Vision-language
LLaVA-NeXT (1.6)
Custom visual assistants and image Q&A
Code under Apache 2.0; weights follow the base LLM license
Vision-language
Florence-2
Captioning, detection, OCR, and segmentation at small scale
MIT
Vision-language
PaliGemma
Compact captioning and visual question answering
Gemma Terms of Use
Embeddings & retrieval
BGE / GTE / E5
Retrieval models for RAG on private corpora
MIT or Apache 2.0, depending on the model
Embeddings & retrieval
Nomic Embed
Open embeddings with long-context support
Apache 2.0
Embeddings & retrieval
Snowflake Arctic Embed
Retrieval-tuned embeddings in several sizes
Apache 2.0
Embeddings & retrieval
Sentence Transformers (MiniLM, MPNet)
Lightweight embeddings for classification and semantic search
Apache 2.0
Embeddings & retrieval
Jina Embeddings
Multilingual retrieval
v2 is Apache 2.0; v3 is non-commercial. Confirm the version
Embeddings & retrieval
ColBERT / rerankers
Late-interaction retrieval and reranking for search quality
MIT or Apache 2.0, depending on the model
Image & video
Stable Diffusion XL
Mature LoRA tooling for brand and product styles
CreativeML Open RAIL++-M
Image & video
FLUX.1 [dev] / [schnell]
High-fidelity text-to-image with LoRA training support
[schnell] Apache 2.0; [dev] non-commercial unless licensed from Black Forest Labs
Image & video
Stable Diffusion 3.5
MMDiT architecture in Medium and Large sizes
Stability AI Community License; free below US$1M annual revenue
Image & video
HiDream-I1
Large open text-to-image model with strong prompt adherence
MIT
Image & video
Qwen-Image
Text-to-image with strong text rendering in images
Apache 2.0
Image & video
Stable Diffusion 1.5
Large legacy LoRA ecosystem with lower hardware needs
CreativeML Open RAIL-M
Image & video
PixArt-Σ / PixArt-α
Efficient text-to-image at modest VRAM
CreativeML Open RAIL++-M
Image & video
Kolors
Chinese and English prompt understanding
Apache 2.0 code; commercial use of weights requires free registration
Image & video
Playground v2.5
Aesthetic-focused model built on the SDXL architecture
Playground v2.5 Community License
Image & video
AuraFlow
Open flow-based text-to-image model
Apache 2.0
Image & video
HunyuanDiT
Diffusion transformer with bilingual prompt support
Tencent Hunyuan Community License; territory restrictions apply
Image & video
AnimateDiff
Motion modules for Stable Diffusion checkpoints
Apache 2.0
Image & video
Wan 2.1
Text-to-video and image-to-video, 1.3B and 14B sizes
Apache 2.0
Image & video
Mochi 1
Open text-to-video model with LoRA fine-tuning support
Apache 2.0
Image & video
CogVideoX
Open text-to-video and image-to-video models
Apache 2.0 (2B); CogVideoX license (5B)
Image & video
ControlNet / IP-Adapter
Pose, depth, edge, and reference-image conditioning
Apache 2.0; the base checkpoint license also applies
Speech
Whisper
Domain-specific speech recognition for calls and meetings
MIT
Speech
Distil-Whisper
Faster, smaller Whisper variants for low-latency transcription
MIT
Speech
wav2vec 2.0 / XLS-R
Speech recognition for low-resource languages and accents
Apache 2.0
Speech
Parakeet / Canary
NVIDIA NeMo speech recognition models
CC-BY-4.0 for Parakeet; Canary terms vary by release