Supported models
You can use any of the following model identifiers as model_id through Cinna’s API or SDK, seamlessly integrated across Solana-based infrastructure.
Language Models
deepseek/deepseek-r1 A high-performance open model rivaling leading proprietary LLMs in math, code, and reasoning
deepseek/deepseek-v3 A massive Mixture-of-Experts model with 685 billion total parameters and state-of-the-art performance
deepseek/deepseek-r1-distill-llama-70b Distilled DeepSeek R1 built on Llama 3 70B for faster and more efficient outputs
NaniDAO/deepseek-r1-qwen-2.5-32B-ablated Qwen-based model designed to avoid refusal logic and always respond with uncensored results
mistralai/mixtral-8x22b-instruct Sparse Mixture-of-Experts model optimized for instruction-following and multilingual interaction
mistralai/mixtral-8x7b-instruct Compact version for cost-effective conversational applications
mistralai/mistral-7b-instruct Lightweight, high-speed language model for real-time deployments
meta-llama/llama-3.3-70b-instruct Latest version of Llama 3 designed for high-quality structured outputs
nvidia/llama-3.1-nemotron-70b-instruct NVIDIA-optimized instruction model tuned for enterprise-grade reasoning and tool use
qwen/qwen-2.5-coder-32b-instruct A leading open-source model for code generation, competitive with proprietary coders
hermes-3-llama3.1-8b Hermes flagship model built for memory, reasoning, and agentic control across long conversations
asi1-mini The first LLM specifically optimized for Solana-native AI agents and Web3 automation
Note: Models marked as uncensored have been finetuned to bypass content refusal logic and always provide output. This is ideal for use cases requiring unrestricted generation.
Image Generation Models
Cinna supports a wide variety of image generation tools across artistic and professional domains:
FLUX.1-dev High-performance open image model suitable for a wide range of styles
Aurora Anime-specific visual model trained for high-detail characters
CyberRealisticXL Ideal for generating realistic portraits and photography-style images
BlazingDrive, BrainDance, YamersCartoonArcadia Tailored for anime, watercolor, and cartoon applications
ArchitectureRealMix Designed for architectural visualization and environmental rendering
ArthemyComics, ArthemyReal Comic and fantasy-style image generation
DreamShaperXL, AAMXLAnimeMix, SDXLUnstableDiffusersV11 Multi-purpose tools covering photography, art, anime, and enhanced diffusion
SDXL The general image generator by Stability AI, fully accessible through Cinna
All models can be used through the standard Cinna API with native Solana-based validation and billing.
Embedding Models
BAAI/bge-large-en-v1.5 A powerful multilingual model for turning natural language into vector embeddings
These embeddings can be used for semantic search, clustering, and memory within Solana AI agent workflows.
Extend the Model Library
If your project requires a specific model not listed here, we support custom deployment from both Civitai and HuggingFace. Whether you want to self-host a model, run it privately on Solana, or deploy a customized version for your use case, Cinna makes it easy to get started.
For support and onboarding, contact the team directly at team@cinna.ai
Let me know if you want this broken into sections for GitBook or displayed in a UI format.
Last updated