Supported models

Every major model, every way you buy it.

Connect model vendors, inference clouds, your own cloud tenancy, self-hosted models and AI subscriptions. Each provider's model list syncs automatically, and every model sits behind the same gateway, the same policy and the same audit trail.

38provider types
4subscription types
8model types
6 hautomatic model sync

Providers

38 ways to connect models.

Each provider type comes with its default address, the country where it processes data and a link to create an API key. Where it runs decides which data your frameworks let it see.

Public cloudPrivate cloudOn-premises

Model vendors 11

Connect the model makers directly with your own API key.

  • OpenAIPublic cloud
  • AnthropicPublic cloud
  • Google GeminiPublic cloud
  • xAI GrokPublic cloud
  • Mistral AIPublic cloud
  • CoherePublic cloud
  • DeepSeekPublic cloud
  • Moonshot AI (Kimi)Public cloud
  • Alibaba Qwen (Model Studio)Public cloud
  • Z.ai (GLM)Public cloud
  • PerplexityPublic cloud

Aggregators & inference clouds 13

Hundreds of open and commercial models through one account each.

  • OpenRouterzero-retention routingPublic cloud
  • Together AIPublic cloud
  • Fireworks AIPublic cloud
  • GroqPublic cloud
  • CerebrasPublic cloud
  • DeepInfraPublic cloud
  • SambaNovaPublic cloud
  • HyperbolicPublic cloud
  • Novita AIPublic cloud
  • Hugging Face InferencePublic cloud
  • NVIDIA NIM (hosted)Public cloud
  • OpenCode (pay-as-you-go)Public cloud
  • ExcelScalePublic cloud

Your cloud tenancy 3

Models in your own cloud account, under your existing agreements.

  • Azure OpenAI / AI FoundryPrivate cloud
  • AWS BedrockPrivate cloud
  • Google Vertex AIPrivate cloud

Self-hosted 7

Models on your own servers and GPUs, for data that must not leave.

  • OllamaOn-premises
  • vLLMOn-premises
  • NVIDIA NIM (self-hosted)On-premises
  • Hugging Face TGIOn-premises
  • LM StudioOn-premises
  • LiteLLM proxyPrivate cloud
  • Any OpenAI-compatible endpointPrivate cloud

AI subscriptions 4direct

Accounts you already pay for, connected directly. Turned on per organisation on request.

  • Claude Pro / MaxPublic cloud
  • ChatGPT / CodexPublic cloud
  • GitHub CopilotPublic cloud
  • Google account (Gemini)Public cloud

Model types

Chat, media, search and safety, all governed.

Every model is typed automatically when it syncs, and each endpoint only routes to models of its type. All of them go through the same detection, policy, guardrails and audit. Which models you can use depends on the providers you connect; these are the families available today.

Chat & reasoning

/v1/chat/completions · /v1/messages · /v1/responses

Assistants, coding agents and reasoning models, with streaming and tool use.

  • GPT & o-series
  • Claude
  • Gemini
  • Gemma
  • Grok
  • Llama
  • Mistral & Ministral
  • Qwen
  • DeepSeek
  • Kimi
  • GLM
  • MiniMax
  • Nemotron
  • Command
  • Sonar
  • Amazon Nova
  • Seed
  • MiMo
  • Hunyuan
  • Ling
  • Step
  • LFM
  • Inkling

Image generation

/v1/images/generations

Create and edit images, with the same checks on the prompt as any other request.

  • GPT Image
  • Gemini image
  • FLUX
  • Recraft
  • Grok Imagine
  • MAI Image
  • MiniMax image

Video generation

/v1/videos

Submit a job, follow its status and download the result through the gateway; the exact cost is recorded when it finishes.

  • Veo
  • Kling
  • Seedance
  • Wan
  • Hailuo
  • Runway Gen-4.5 · Aleph
  • Grok Imagine Video
  • FLUX video
  • Gemini Omni
  • HeyGen

Embeddings

/v1/embeddings

Vectors for search and retrieval-augmented generation.

  • OpenAI text-embedding-3
  • Gemini Embedding
  • Cohere Embed
  • Mistral & Codestral Embed
  • Qwen3 Embedding
  • Nemotron Embed
  • Amazon Titan
  • Perplexity Embed
  • EmbeddingGemma
  • BGE
  • E5
  • GTE

Text to speech

/v1/audio/speech

Turn text into natural speech.

  • OpenAI TTS
  • Gemini TTS
  • Grok TTS
  • Voxtral TTS

Rerank

/v1/rerank

Put search results in order of relevance before they reach a model.

  • Cohere Rerank
  • Voyage Rerank
  • BGE Reranker

Moderation

/v1/moderations

Classify content against safety categories.

  • OpenAI omni-moderation

Guardrail models

used inside governance profiles

Safety models that screen requests and responses for your profiles, ideally on your own hardware.

  • Llama Guard 3
  • ShieldGemma
  • Granite Guardian
  • any safe/unsafe classifier

What models can take in

Images, PDFs, audio and video, where the model supports them.

Each model is marked with what it accepts, read from the public model catalogue during sync or from well-known model families. The AI workspace uses this to offer attachments only for models that can handle them.

Data detection reads text, so profiles can block images, audio, video and files outright for sensitive workloads. Endpoints whose content can't be inspected before it leaves, such as audio transcription, batch and file uploads, are refused rather than passed through.

  • Visionimages in the prompt
  • PDFdocuments in the prompt
  • Audio inaudio in the prompt
  • Video invideo in the prompt
  • Image outgenerates images
  • Audio outgenerates speech
  • Video outgenerates video
  • Guardrailscreens other requests

How models stay current

New models appear on their own. Retired ones step aside.

Synced automatically

When a provider is added, when its key changes, every six hours, and whenever you press "Sync models".

Read from the provider

Each provider's own model list, using its stored key. Prices and context windows come from the public model catalogue.

Disabled, never deleted

A model the provider stops listing is disabled, so its history, rules and reports stay intact.

Ready for rules

OpenRouter models with zero-retention routes are marked, kept current hourly, so rules like "zero-retention providers only" pick the right models.

How rules choose between models →

Connect the models you already use.

Add a provider or subscription in the console and its models appear automatically, governed from the first request.