Supported Models
Tangle Inference reaches every supported provider through a single API. Which models serve right now changes when a provider account refuses or recovers, so this page names none of them. The live catalog does: GET /v1/models with your API key.
What serves now
Every catalog row carries an availability object:
| Field | Meaning |
|---|---|
status | available (served recently), unknown (no recent request to judge by), or unavailable (refused right now) |
reason, message | Why a row is unavailable, for example provider_quota_exhausted |
observed_at | When the state was observed; for an unavailable row, when the refusal began |
last_served_at, served_by | When the model last answered, and through which provider |
credit_input_modalities | The input kinds a credit-paid key can send to this model: text, plus image where its provider documents a per-image token bound. Set an agent’s input kinds from this, not from architecture.input_modalities |
By default the catalog lists available and unknown rows. Add ?availability=all to see every row with its reason, or ?availability=available for only the models that answered recently.
A request for a model whose every route is refused gets 503 provider_quota_exhausted at once, with no upstream call. The error names the provider, when the refusal began and, when one exists, a model of the same family that serves now.
Plus operators registered on Tangle and running Blueprints:
| Blueprint | Models | How to route |
|---|---|---|
| LLM Inference | Llama, Qwen, Mistral, any vLLM-compatible model | X-Tangle-Routing: operator or auto |
| Vector Store | Embedding models for RAG | Pin by Blueprint ID or operator slug |
| Custom Blueprints | Any model the operator deploys | Pin by Blueprint ID or operator slug |
Operators set their own pricing and the gateway scores them on reputation, latency, and price. See Operator Routing for details.
Model ID format
Use provider/model-name:
openai/gpt-4o-mini
google/gemini-2.5-flash
deepseek/deepseek-v4.1-flashOr use bare names: the gateway resolves the provider by prefix. A resolved provider can still be refusing; check the row’s availability.
| Prefix | Resolves to |
|---|---|
gpt-, o1-, o3-, o4- | OpenAI |
claude- | Anthropic |
gemini-, gemma- | |
llama-, mixtral- | Groq |
deepseek- | DeepSeek |
mistral-, codestral- | Mistral |
grok- | xAI |
glm- | Z.ai |
command- | Cohere |
Modalities
| Modality | Endpoint | Examples |
|---|---|---|
| Text | /v1/chat/completions | All chat models |
| Images | /v1/images/generations | DALL-E, FLUX |
| Audio | /v1/audio/transcriptions, /v1/audio/speech | Whisper, TTS |
| Embeddings | /v1/embeddings | text-embedding-3-small/large |
| Video | /v1/videos/* | Avatar generation, dubbing |
Model catalog API
The model catalog is available at GET https://router.tangle.tools/v1/models with your API key, with pricing, context length, and modality information for every model.