Skip to main content

Providers

The AI Gateway supports 15 third-party AI providers with 268 certified models, plus self-hosted models. All providers use a unified API interface. The models listed here are each vendor's current models (October 2026); models a vendor retires leave the catalog (see Certified Models).

Provider Summary​

ProviderModelsCapabilities
OpenAI109Chat, Embeddings, TTS, STT, Realtime, Images
Anthropic14Chat with Vision
Google (Gemini)35Chat, Embeddings, TTS, Images, Video
Mistral33Chat, Embeddings, STT, OCR, Moderation
Cohere25Chat, Embeddings, Reranking
Grok (xAI)11Chat, Images
DeepSeek2Chat, Reasoning
ElevenLabs13TTS, Speech-to-Speech, Music, Sound Effects
Mubert-Music Generation
Deepgram-Speech-to-Text
Stability AI8Image Generation
Black Forest Labs11Image Generation (FLUX)
Runway5Video Generation
Luma AI2Video Generation
Custom Provider-Any OpenAI-compatible API
Self-Hosted169Self-hosted models (engine determines protocol: vllm, transformers, whisper, tts-engine, video-engine, custom)

Provider API Keys and Adding a Model​

Save a provider's API key under AI Gateway > API Keys, choosing the provider from the same cards as the Third Party AI page. Test checks a saved key by listing the provider's models with it; no model is called, so the test does not depend on any one model being available.

On the Third Party AI page, pick a provider card and choose the model from the provider's certified models (the field shows a current model of that provider as its example, such as gpt-6.1-sol for OpenAI or claude-opus-5-5 for Anthropic). Test Connection sends that model a minimal request with your key.

OpenAI​

Models for chat, reasoning, embeddings, audio, realtime voice, and image generation.

Capabilities: Chat, Embeddings, Text-to-Speech, Speech-to-Text, Realtime, Image Generation

ModelTypeBest For
openai/gpt-6-astraChatOpenAI's most capable model, for the most demanding work
openai/gpt-6.1-solChatComplex coding and professional tasks at lower cost than Astra
openai/gpt-6-lunaChatEfficient, high-volume tasks
openai/text-embedding-3-largeEmbeddingHigh-quality embeddings
openai/gpt-4o-mini-ttsTTSSpeech synthesis
openai/gpt-transcribeSTTAudio transcription
openai/gpt-realtime-2.1RealtimeVoice agents with reasoning and tools
openai/gpt-image-2.5-sunburstImageHighest-quality image generation
openai/gpt-image-2.5-flareImageFast everyday image generation

Anthropic​

Claude models with strong reasoning, coding, and long-context capabilities (1M-token context window).

Capabilities: Chat with Vision, Function Calling, Web Search

ModelTypeBest For
anthropic/claude-opus-5-5ChatLong-running agentic coding and knowledge work
anthropic/claude-fable-5-1ChatDemanding reasoning and long-horizon agentic work
anthropic/claude-sonnet-5-5ChatBest combination of speed and intelligence
anthropic/claude-haiku-5-5ChatHigh-volume, latency-sensitive tasks

Google (Gemini)​

Multimodal models with large context windows and diverse capabilities.

Capabilities: Chat, Embeddings, TTS, Image Generation, Video Generation

ModelTypeBest For
gemini/gemini-3.8-flashChatFast multimodal processing with search grounding
gemini/gemini-3.1-pro-previewChatMost capable Gemini model
gemini/gemini-3.5-flash-liteChatFast, cost-effective
gemini/gemini-embedding-2EmbeddingText embeddings
gemini/gemini-3.8-flash-ttsTTSSpeech synthesis
gemini/gemini-nano-banana-2.1ImageImage generation and editing
gemini/veo-3.1-generate-previewVideoVideo generation

Note: In the backend provider enum, Google models use the gemini provider identifier. When configuring models via the API, use gemini as the provider value.

Mistral​

European AI with strong multilingual and code capabilities.

Capabilities: Chat, Embeddings, Speech-to-Text, OCR, Moderation

ModelTypeBest For
mistral/mistral-large-4ChatMost capable Mistral model
mistral/mistral-medium-2604ChatMistral Medium 3.5, balanced quality and cost
mistral/mistral-small-2603ChatMistral Small 4, fast and cost-effective
mistral/codestral-latestChatCode generation
mistral/mistral-embedEmbeddingText embeddings
mistral/mistral-ocr-4-1OCRDocument text and layout extraction

Cohere​

Enterprise-focused models with RAG and reranking specialization.

Capabilities: Chat, Embeddings, Reranking

ModelTypeBest For
cohere/command-a-plus-05-2026ChatCohere's most capable model
cohere/command-a-reasoning-08-2025ChatReasoning tasks
cohere/command-a-vision-07-2025ChatVision tasks
cohere/embed-v5.0-proEmbeddingMultilingual, multimodal embeddings
cohere/embed-v5.0-fastEmbeddingFaster multimodal embeddings
cohere/rerank-v4.0-proRerankSearch result reranking

Grok (xAI)​

xAI's models with real-time knowledge and image generation.

Capabilities: Chat, Vision, Image Generation

ModelTypeBest For
grok/grok-4.7ChatMost capable Grok model, code and chat
grok/grok-4.3Chat1M-token context window
grok/grok-4.20-0309-non-reasoningChatFast answers without reasoning
grok/grok-build-0.1ChatCoding
grok/grok-imagine-image-qualityImageImage generation

DeepSeek​

Strong reasoning and coding capabilities. Thinking is on when a request sets a reasoning effort.

Capabilities: Chat, Reasoning

ModelTypeBest For
deepseek/deepseek-v4-proChatComplex reasoning and coding
deepseek/deepseek-flashChatFast, cost-effective chat

ElevenLabs​

Voice synthesis and audio generation.

Capabilities: Text-to-Speech, Speech-to-Speech, Music Generation, Sound Effects

ModelTypeBest For
elevenlabs/eleven_v4TTSMost expressive speech, with audio tags
elevenlabs/eleven_v4_turboTTSFast expressive speech
elevenlabs/eleven_multilingual_v2TTSMultilingual voice synthesis
elevenlabs/eleven_flash_v2_5TTSUltra-fast TTS
elevenlabs/eleven_musicMusicMusic generation
elevenlabs/sound-effects-v2AudioSound effects

Mubert​

AI-powered music generation platform.

Capabilities: Music Generation

ModelTypeBest For
mubert/mubert-ttmMusicAI-generated music and soundtracks

Deepgram​

Speech-to-text.

Capabilities: Speech-to-Text

ModelTypeBest For
deepgram/nova-3STTGeneral-purpose transcription

Stability AI​

Stable Diffusion image generation models.

Capabilities: Image Generation

ModelTypeBest For
stability/stable-image-ultraImageHighest-quality image generation
stability/sd3.5-largeImageHigh-quality image generation
stability/sd3.5-large-turboImageFast image generation

Black Forest Labs​

FLUX models for image generation.

Capabilities: Image Generation

ModelTypeBest For
bfl/flux-2-maxImageHighest-quality FLUX.2 images
bfl/flux-2-proImageProfessional quality
bfl/flux-2-flexImageAdjustable quality and speed
bfl/flux-2-klein-9bImageFast generation
bfl/flux-kontext-proImageImage editing with context

Runway​

Video generation for creative applications.

Capabilities: Video Generation

ModelTypeBest For
runway/gen4.5VideoText or image to video
runway/gen4_turboVideoFast image to video
runway/aleph2VideoVideo editing (video to video)

Luma AI​

AI video generation with Ray models.

Capabilities: Video Generation

ModelTypeBest For
luma/ray-2VideoHigh-quality video
luma/ray-flash-2VideoFast video generation

Self-Hosted​

Self-hosted models deployed on your own infrastructure. 169 pre-configured models are available across all AI capabilities.

Provider: self_hosted (always -- the engine field determines the API protocol)

Engines:

EngineModelsUsed For
vllm113LLM chat, multimodal, embedding, image generation
transformers24Custom HuggingFace models not supported by vLLM
tts-engine16Text-to-speech models
whisper5Speech-to-text models
video-engine6Video generation models
custom5Generic custom inference servers

Capabilities by Engine:

Capabilityvllm/transformerswhispertts-enginevideo-enginecustom
Chat completionsYes---Yes
StreamingYes---Yes
EmbeddingsYes---Yes
Vision/MultimodalYes----
Text-to-Speech--Yes-Yes
Speech-to-Text-Yes--Yes
Image GenerationYes---Yes
Video Generation---YesYes
Tool CallingYes----

See Self-Hosted Models for the full engine architecture and model catalogue.

Using Models​

When calling the AI Gateway, use the Strongly-generated model ID - the unique identifier assigned when a model is added to your account.

// Get your model ID from your configured models list
const modelId = '507f1f77bcf86cd799439011';

const response = await fetch('https://ai-gateway.strongly.ai/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'X-User-Id': userId,
    'X-App-Id': appId  // or X-Workflow-Id or X-Workspace-Id
  },
  body: JSON.stringify({
    model: modelId,  // Strongly-generated model ID
    messages: [{ role: 'user', content: 'Hello!' }]
  })
});

Add web_search_options to a chat request and the model searches the web before it answers, citing the pages it used. Send {} for the defaults, or {"max_uses": 3} to cap how many searches a request runs.

body: JSON.stringify({
  model: modelId,
  messages: [{ role: 'user', content: 'What is the latest stable Python release?' }],
  web_search_options: {}
})

Which search a model uses depends on the model:

ModelsSearch
OpenAI: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna, GPT-5 through GPT-5.6 (with their mini, nano and pro models, GPT-5.3 Codex and the GPT-5 search models), GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini, o3 and o3-proOpenAI's own web search
Anthropic: Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1 and every earlier Claude model still served (Haiku 4.5, Sonnet 4.5 and later, Opus 4.5 and later, Fable 5)Anthropic's own web search
Google: Gemini 3.8, 3.7, 3.6 and 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 3 Flash, 3.1 Pro, Robotics-ER 2, 2.5 Flash, 2.5 Pro, 2.5 Flash-Lite, and the flash-latest, flash-lite-latest and pro-latest aliasesGoogle Search grounding
Every other model that can call tools: other providers, self-hosted models, and OpenAI models without a search of their own (the Codex and chat-latest models)Strongly's search, run by the AI Gateway

A model that can neither search nor call tools (for example an audio-only model) is refused a web search request with the reason; it never answers as if it had searched.

An OpenAI reasoning model asked to search at minimal reasoning effort searches at low, the lowest effort OpenAI's search runs at.

Switching Providers​

To switch between providers, use the Strongly ID for the model you want to use:

// Each model has its own unique Strongly ID
const openaiModel = '507f1f77bcf86cd799439011';     // Your GPT-6.1 Sol instance
const anthropicModel = '507f1f77bcf86cd799439012';  // Your Claude instance
const googleModel = '507f1f77bcf86cd799439013';     // Your Gemini instance

// Use any model by passing its Strongly ID
const model = openaiModel;

Note: The same vendor model (e.g., GPT-6.1 Sol) can be configured multiple times with different API keys. Each configuration gets a unique Strongly ID.