Supported Models
Ghast doesn't ship with a fixed model list. Models come from the providers you enable — and what's available depends on what each provider declares.
Where the list comes from
| Source | What it adds |
|---|---|
| 0G Router | A curated set of models served by the Router endpoint. Updated periodically by the Router operator. |
| 0G Advanced (on-chain registry) | Models declared by each enabled Advanced provider in their on-chain service entry. |
| Plugins | Models registered by installed plugins (e.g. a local Ollama plugin, a custom OpenAI-compatible plugin). |
The desktop client refreshes the lists periodically and on demand (when you open the model picker). For 0G Advanced, the list is read directly from the chain.
Capability declaration
Each model exposes capability data that Ghast turns into picker badges:
| Field | Where it lives | Badge implications |
|---|---|---|
supportedParameters: ["stream"] | Provider response | Streaming badge |
supportedParameters: ["tools"] or ["tool_choice"] | Provider response | Tools badge — required for skills, MCP, Computer Use |
architecture.input_modalities.includes("image") | On-chain (Advanced) or response | Vision badge |
architecture.output_modalities.includes("image") | On-chain (Advanced) or response | Image-output badge |
supportedParameters: ["response_format"] | Provider response | JSON badge |
| Name / description matching reasoning patterns | Either | Reasoning badge |
| Advanced provider (any) | On-chain | TEE badge |
If a model lacks a capability you need, the composer refuses to attach the corresponding input. For example, attaching an image to a non-vision model is blocked at the composer level, not silently dropped.
Why no fixed list
There are a few reasons Ghast doesn't pin a static model list in the documentation:
- Routers update. What
0G Routerserves today may not be what it serves in three months. The picker is the source of truth. - Advanced providers are on-chain. The set is governed by the chain registry, not the client.
- Plugins are user-installed. Ghast doesn't know what you'll add.
So this page describes how to read the picker, not what's in it. For the live list, open the model picker.
Pricing per Advanced provider
Each Advanced provider declares pricing on-chain:
- Output token price
- (Sometimes) input token price
- Floor amount per transfer
Funding flows respect both the per-model floor and the global MIN_TRANSFER_AMOUNT(). See Inference Modes for the funding logic.
The picker doesn't show pricing inline; for cost comparison you can check Settings → Providers → 0G Advanced → [provider] → Pricing.
Models the agent can use vs. the ones you can use
A few capabilities are gated:
| Feature | Required model capability |
|---|---|
| Skills, tools, MCP | Tools badge |
| Computer Use | Tools badge |
| Image attachments | Vision badge |
| Mermaid / diagram rendering | (Client-side, not model) |
| Structured generation | JSON badge |
| Extended thinking | Reasoning badge |
If you pick a model without one of these, the agent loses the capability for that turn. The composer disables the corresponding controls.