Models
The QDivZero model catalog
Start with one. Deploy almost any model.
Use QDivZero public models directly or explore open-weight Hugging Face models for text, code, reasoning, image, video, audio, embeddings, and multimodal applications.
Need another one? Deploy it on dedicated compute. Is the model yours? You can also run private and fine-tuned models. QDivZero prepares the infrastructure and inference environment so you can connect through an OpenAI-compatible API.
Explore modelsPublic models
Start building without deploying anything.
To test a model, develop a prototype, or start integrating it into your application, you do not always need to create a deployment.
QDivZero public models are ready to use through the API and deduct balance according to input and output tokens. Qwen 3.8 27B can also be activated with Unlimited Basic for a fixed monthly fee without using balance. Connect through an OpenAI-compatible API and start sending requests without configuring GPUs, servers, or inference environments.
Use models such as DeepSeek V4 Flash, DeepSeek V4.1 Flash, GPT-6 Luna, GPT-6 Sol, GPT-OSS Safeguard 120B European, GPT-OSS Safeguard 20B European, and Qwen 3.8 27B, with input and output pricing visible before you start.
View public modelsOpen-weight models
You found it on Hugging Face. Now run it.
Finding an open-weight model is easy. Running it means knowing what hardware it needs, finding a GPU with enough VRAM, preparing the runtime, configuring inference, and keeping an endpoint available.
QDivZero connects that model with the infrastructure required to run it. Find the Hugging Face model you want to deploy, define your requirements, and QDivZero validates compatibility, finds suitable capacity, and prepares the inference environment.
You do not have to wait for us to add the model to a closed list. If it is compatible, you can deploy it.
Explore Hugging Face modelsAn open catalog
Do not wait for your provider to add the next model.
Open-weight models move quickly. New architectures, versions, and specialized models appear continuously, while many APIs rely on a selection that the provider adds and maintains manually.
QDivZero starts from the Hugging Face ecosystem so you can explore new models without depending on that selection. From large language and reasoning models to vision, image and video generation, speech, transcription, or embeddings.
You decide which model is worth testing. Not your provider's catalog.
Your own models
Your model has a place here too.
Not everything has to come from a public catalog. If you developed your own model, fine-tuned an existing one, or need to keep its weights private, you can deploy it on dedicated compute in QDivZero.
Run custom, private, and fine-tuned models without manually building the inference infrastructure. QDivZero prepares the required environment and keeps the model available through an endpoint you can integrate into your application.
Your weights remain yours. Infrastructure stops being your problem.
Deploy my modelRight now
Some of our favorite models.
There is no single model that is right for everything. These are some of the open-weight models worth exploring depending on what you are building.
Qwen3.8-27B combines capability with a manageable size for inference. DeepSeek V4.1 and GLM-5.3 target reasoning and more complex tasks. Among MoE architectures, Ornith-35B-A3B, Qwen3.6-35B-A3B, and Sherry-35B-A3B let you work with a fraction of the parameters active during inference.
For agents, context, and tool use, explore Kimi K3 and MiniMax M2.7. MiMo-V2.6-Flash and MiMo-V2.6-Pro offer different speed and capability profiles, while Gemma 4 31B expands the options for multimodal applications and NVIDIA Nemotron 3.5 for reasoning and agents.
Find these models and many more in QDivZero. Try them when they are available as public models or deploy them on dedicated compute without managing the GPU infrastructure yourself.
Explore all modelsFrom model to endpoint
Deploy AI models without configuring the infrastructure.
Choose the model you want to run and define your requirements. QDivZero checks the runtime and VRAM it needs, searches for compatible compute across providers, and prepares the inference environment.
When the deployment is ready, connect through an OpenAI-compatible API. Change models or infrastructure without building a different integration for every provider.
Create a deploymentPopular models
Selection updated from Hugging Face
Generative model for creating images from text instructions.
Open-weight model for text generation, reasoning, and assistants.
Open-weight model for text generation, reasoning, and assistants.
Open-weight model for text generation, reasoning, and assistants.
Multimodal model for understanding images and generating responses.
Multimodal model for understanding images and generating responses.
Multimodal model for understanding images and generating responses.
Open-weight model for text generation, reasoning, and assistants.
Open-weight model for text generation, reasoning, and assistants.
Generative model for turning images into video sequences.
Audio model for transcribing speech into text.
Generative model for creating audio from text instructions.
Open-weight model for text generation, reasoning, and assistants.
Multimodal model for understanding images and generating responses.
Open-weight model for text generation, reasoning, and assistants.
Generative model for creating images from text instructions.
Audio model for detecting and separating speech activity.
Multimodal model for generating video from image and text.
Model for analysing, labelling, and classifying text content.
Multimodal model for understanding images and generating responses.
QDivZero public models
Ready to use from the platform
public:qwen-3.8-27b
Dense 27B multimodal model for coding, reasoning, agents, and image and video understanding, with flexible reasoning control.
per 1M tokens
public:deepseek-v4-flash-0731
Official DeepSeek V4 Flash release for agents, coding, and tool use, with speculative decoding to accelerate inference.
per 1M tokens
public:deepseek-v4.1-flash
DeepSeek multimodal model for understanding text and images, with an architecture optimized to reduce context memory during inference.
per 1M tokens
public:gpt-6-luna
Efficient OpenAI model for focused, high-volume tasks, with reasoning, text and image input, and tool use.
per 1M tokens
public:gpt-6-sol
OpenAI model balancing capability and efficiency for complex coding, reasoning, and agentic workflows.
per 1M tokens
public:gpt-oss-safeguard-120b-european
117B-parameter safety reasoning model for classifying content against custom policies, filtering inputs and outputs, and explaining decisions.
per 1M tokens
public:gpt-oss-safeguard-20b-european
Compact, lower-latency model for safety classification, policy-based moderation, and content labeling.
per 1M tokens
Frequently asked questions
Does QDivZero include the entire Hugging Face catalog?
Practically all of it. Instead of limiting you to a selection of models, QDivZero gives you access to the Hugging Face catalog to explore and deploy compatible models. Only a small portion may be excluded because of technical or architectural requirements. If it is compatible, it is available. You do not have to wait for us to decide to add it.
Which Hugging Face models can I deploy on QDivZero?
You can deploy compatible models for text, reasoning, image, video, audio, speech, vision, embeddings, and multimodal tasks, among many others available on Hugging Face. From large language models to small, specialized models, QDivZero checks their requirements and finds the right infrastructure to run them.
What is the difference between QDivZero and a model API provider?
An API provider usually offers a specific selection of pre-integrated models. QDivZero does not start from a closed list:it starts with the Hugging Face catalog. This lets you choose the model first and let QDivZero solve the compatible infrastructure required to run it.
Can I deploy new models published on Hugging Face?
Yes. If a new model appears on Hugging Face and is compatible with QDivZero, you can deploy it without waiting for us to manually add it to a list. This also gives you access to new and specialized models that may not be available in other providers' usual catalogs.
Can I deploy LLMs and open-weight models on QDivZero?
Yes. You can deploy compatible LLMs and other open-weight models for chat, reasoning, text generation, agents, and other AI applications. The catalog includes model families such as Qwen, DeepSeek, GLM, Kimi, MiMo, MiniMax, Gemma, and NVIDIA Nemotron, along with many others available on Hugging Face.
Is QDivZero only for text models?
No. You can deploy models for text, image, video, audio, speech, vision, embeddings, and multimodal applications. This makes QDivZero useful for much more than LLMs: from image and video generation to transcription, semantic search, RAG, and applications that combine different modalities.
Do I need to know which GPU a Hugging Face model needs?
No. You choose the model and QDivZero finds compatible capacity to run it. You do not need to research which GPU it needs, check availability, contract infrastructure with another provider, or configure the inference environment manually.
How do I use a model after deploying it?
Once deployed, you get an inference endpoint that you can connect to your application. Compatible models can be used through the QDivZero OpenAI-compatible API, making integration easier for applications, agents, and tools that already work with this format.
How much does it cost to run a Hugging Face model on QDivZero?
With QDivZero, you can use balance, an Unlimited Basic subscription, or both in the same account. When you deploy a Hugging Face model, active compute is prorated by the second and deducted from your balance. Public Models deduct balance according to input and output tokens, while Unlimited Basic provides Qwen 3.8 27B for a fixed monthly fee without using balance. You can see the compute price before deploying and choose the payment method that best fits each workload.
What are QDivZero public models?
They are models that QDivZero already keeps deployed and ready to use through the API. You do not need to create a deployment, select a GPU, or prepare an inference environment before sending your first request. Public Models use the OpenAI-compatible API and deduct balance according to input and output tokens. Qwen 3.8 27B can also be used with Unlimited Basic for a fixed monthly fee without using balance.
What is the difference between a public model and a dedicated deployment?
A Public Model is ready to use and deducts balance according to input and output tokens. A dedicated deployment reserves infrastructure for the model you choose and deducts balance for active compute time. Unlimited Basic includes Qwen 3.8 27B as a third option with a fixed monthly fee that does not use balance. A dedicated deployment is the right option when you want to run a different open-weight model, control its infrastructure, or use a custom private model.
Can I deploy my own private or fine-tuned models?
Yes. You can run custom, private, or fine-tuned models on dedicated compute. QDivZero prepares the required runtime and inference infrastructure and keeps the model available through an endpoint for your application. This lets you use your weights without making the model part of a public catalog.
Why deploy Hugging Face models with QDivZero instead of doing it yourself?
Running a model yourself means finding compatible infrastructure, choosing GPUs, checking availability, configuring the inference environment, and maintaining the deployment. With QDivZero, you start with the model. We solve the infrastructure required to run it, so you can move from Hugging Face to an endpoint without managing servers or GPUs.