Deployable
Gemma 4 26B A4B
- Text and chat
- Apache 2.0
Mixture-of-experts model with vision input, about 4B parameters active per token.
24 GB GPU or more
- License
- Apache 2.0
- Runs with
- Ollama (4-bit)
- Weights
- google/gemma-4-26B-A4B-it
AI Studio
Run AI modelsOn rented GPUs.
Open-weight models whose licenses allow paid hosting, deployed to a 24 GB GPU on the Nosana network and paid from your own wallet. You get a private endpoint and a playground; stop any time and unused time comes back.
12 models
4 deployable today
24 GB GPUs
Premium markets
One transaction
Fee shown first
Pay with Credits
Top up with SOL
AI models
Workspace
Models
Showing 12 of 12 models. Every one is open-weight, under Apache 2.0, MIT or OpenRAIL++-M.
Google
- Deployable
gpt-oss-20b
OpenAI
- Text and chat
- Apache 2.0
Reasoning model with a 131K context window that runs on a single 16 GB card.
16 GB GPU or more
- License
- Apache 2.0 plus gpt-oss usage policy
- Runs with
- Ollama or vLLM
- Weights
- openai/gpt-oss-20b
- Deployable
Qwen3.5 9B
Qwen
- Text and chat
- Apache 2.0
Small multimodal generalist with a 262K context window.
24 GB GPU or more
- License
- Apache 2.0
- Runs with
- Ollama or vLLM
- Weights
- Qwen/Qwen3.5-9B
- Deployable
Stable Diffusion XL 1.0
Stability AI
- Images
- CreativeML OpenRAIL++-M
Image model with the largest LoRA and ControlNet ecosystem.
12 GB GPU or more
- Runs with
- ComfyUI
- Coming next
FLUX.2 [klein] 4B
Black Forest Labs
- Images
- Apache 2.0
Text-to-image and image editing in four steps.
24 GB GPU or more
- License
- Apache 2.0
- Runs with
- diffusers
Deploys after a test on real nodes.
- Coming next
Whisper large-v3 turbo
OpenAI
- Speech
- MIT
Speech to text in many languages, through faster-whisper.
8 GB GPU or more
- License
- MIT
- Runs with
- faster-whisper
Deploys after a test on real nodes.
- Catalog
BGE-M3
BAAI
- Embeddings
- MIT
Multilingual dense and sparse embeddings for search and retrieval.
8 GB GPU or more
- License
- MIT
- Runs with
- Text Embeddings Inference
- Weights
- BAAI/bge-m3
Not offered for deploys yet.
- Catalog
gpt-oss-120b
OpenAI
- Text and chat
- Apache 2.0
The larger gpt-oss model. Fits one 80 GB card.
80 GB GPU or more
- License
- Apache 2.0 plus gpt-oss usage policy
- Runs with
- Ollama or vLLM
- Weights
- openai/gpt-oss-120b
Not offered for deploys yet.
- Catalog
Kokoro 82M
hexgrad
- Speech
- Apache 2.0
Small text-to-speech model.
8 GB GPU or more
- License
- Apache 2.0
- Runs with
- Kokoro FastAPI
- Weights
- hexgrad/Kokoro-82M
Not offered for deploys yet.
- Catalog
Qwen3.8 27B
Qwen
- Text and chat
- Apache 2.0
Dense 27B model with vision input, served in FP8.
48 GB GPU or more
- License
- Apache 2.0
- Runs with
- vLLM
- Weights
- Qwen/Qwen3.8-27B
Not offered for deploys yet.
- Catalog
Wan2.2 TI2V 5B
Wan-AI
- Video
- Apache 2.0
Text and image to 720p video on one card. Time per clip still has to be measured on Nosana nodes.
24 GB GPU or more
- License
- Apache 2.0
- Runs with
- diffusers
- Weights
- Wan-AI/Wan2.2-TI2V-5B
Not offered for deploys yet.
- Catalog
Z-Image Turbo
Tongyi-MAI
- Images
- Apache 2.0
6B image model with photoreal output and readable text in images.
24 GB GPU or more
- License
- Apache 2.0
- Runs with
- diffusers
- Weights
- Tongyi-MAI/Z-Image-Turbo
Not offered for deploys yet.
Quick deploy
Loading live GPU prices…
Your deployments
View allNo deployments from this browser yet. A deployment shows here as soon as you sign it.
How a deploy works
- Choose a model and a GPU market that has room for it.
- Choose how long to reserve. The page shows the on-chain price plus the Gridora fee (5%).
- Sign one transaction in your wallet. The deposit sits in the Nosana program, not with Gridora.
- Use the private endpoint or the built-in playground while the job runs.
- Stop whenever you like. Unused time goes back to your wallet.
Left out on purpose
- Closed models such as GPT, Claude, Gemini, Sora and Runway. Their weights are not published, so they can't run on a rented GPU.
- Non-commercial licenses such as FLUX.1 [dev], FLUX.2 [dev] and Qwen-Image-2.1. Hosting them for paying users needs a separate license from their makers.
- Licenses with revenue caps or region limits, such as Llama, Stable Diffusion 3.5 and HunyuanVideo, until each user can accept those terms at checkout.
- Models too large for one GPU, such as DeepSeek V3. Multi-GPU jobs come later.
You are responsible for what you generate and how you use it, under each model's license and usage policy and the acceptable use policy. Gridora claims no rights in your outputs.