One integration. Many supported models.

Integrate once. Use the model that fits.

Connect through one authenticated API—or start with the official Go, Python, and command-line clients. Then route canonical messages to any supported text provider and model without adding another provider SDK, credential flow, or request contract to your product.

Start integrating Explore supported models

Keep one canonical request contract as models change
Start with HTTP, Go, Python, or the command line
Keep upstream credentials out of client requests

One endpointRoute across native and compatible provider APIs.

One credentialApplications send a tenant secret, never provider keys.

One contractConsistent request, error, usage, and timeout behavior.

Your product HTTP · Go · Python · CLI

LLM Proxy Authenticate · validate · route

Choose a model family

Choose an exact model

Claude Fable2 exact models

Choose a provider offering

Provider offerings1 route

Supported tasks

Generate textRequired input: Text. Output: Text.
Understand imagesRequired input: Text, Image. Output: Text.

Integrate once

Use the API directly or start with a client.

Official clients use canonical POST /v2. OpenCode and OpenAI clients use the /v1 client protocols. Each protocol uses the same provider coordinator.

1
Choose an integration
Use ordinary HTTP or an official Go, Python, or command-line client.
2
Authenticate once
Give the application one client key while provider credentials remain behind the proxy.
3
Select the model
Use centrally managed defaults or choose a supported provider and model for one request.

OpenCode and OpenAI clients

Connect with a base URL and client key.

Use Chat Completions or Responses with an exact provider/model identifier. Caller tools execute in your client.

Read the client API reference

HTTP API

Any language. No client dependency.

Send canonical JSON over HTTPS and keep provider-specific wire formats out of application code.

Read the v2 reference

Official Go client

Typed requests and media attachments.

Construct validated messages, attach supported media, and preserve typed transport and HTTP failures.

Use the Go client

Official Python client

A small synchronous client.

Send the same canonical messages with explicit configuration and provider/model selection.

Use the Python client

Official CLI

Pipe prompts from scripts and tools.

Install the reusable command for stdin workflows, output limits, and request-level time budgets.

Install the CLI

Built for the way teams ship

One boundary. Three ways to benefit.

Start with the path closest to your work. Each path reaches the same API, client-key boundary, and supported model catalog.

AI-assisted builders

Turn generated code into a durable integration.

Give coding agents and rapid prototypes one documented request shape instead of accumulating provider SDKs and raw keys.

Start with a copyable request

Startups and product teams

Test model fit without multiplying integrations.

Compare supported providers behind the same product boundary, then change the route as quality, capability, or product needs evolve.

Compare native providers

Platform and engineering teams

Standardize access across applications.

Centralize provider credentials, routing validation, usage signals, and client access while product teams keep one contract.

Design an internal gateway

What stays stable

Change providers without rebuilding the product boundary.

LLM Proxy owns provider-specific wire contracts, credential handling, route validation, and observable failure behavior so application code can stay focused on the job it performs.

Switch routes, not integrations.

Select a supported provider and model per request, use managed defaults, or update an application-owned model profile without replacing the client contract.

Keep secrets server-side.

Tenant-secret authentication gives each application access without exposing upstream provider keys. Client access can be generated and rotated independently, while managed provider keys are encrypted at rest and explicitly verified before use.

Explore the security boundary

Keep one request and error contract.

Send canonical messages while the proxy maps provider payloads, output limits, timeouts, usage metadata, and provider failures into one documented boundary.

Inspect the API contract

Use capabilities the route declares.

Send images or audio, request reasoning or constrained web search, or transcribe audio only when the selected model explicitly publishes that capability.

Inspect model capabilities

Current runtime contract

Providers and model capabilities.

This matrix is generated from the same validated provider registry used by request routing. It describes proxy support—not whether a specific account has configured a provider key.

17Providers

15Publishers

33Families

86Exact models

89Offerings

Provider-independent exact models and every current provider offering.

Anthropic`anthropic`	`claude-fable-5`Claude Fable · claude-fable-5	Anthropic`anthropicanthropic_messages`128000 token output
Anthropic`anthropic`	`claude-fable-5-1`Claude Fable · claude-fable-5-1	Anthropic`anthropicanthropic_messages`Reasoning: low, medium, high, xhigh, max128000 token output
Anthropic`anthropic`	`claude-haiku-4-5`Claude Haiku · claude-haiku-4-5	Anthropic`anthropicanthropic_messages`64000 token output
Anthropic`anthropic`	`claude-haiku-4-5-20251001`Claude Haiku · claude-haiku-4-5-20251001	Anthropic`anthropicanthropic_messages`64000 token output
Anthropic`anthropic`	`claude-opus-4-8`Claude Opus · claude-opus-4-8	Anthropic`anthropicanthropic_messages`128000 token output
Anthropic`anthropic`	`claude-opus-5`Claude Opus · claude-opus-5	Anthropic`anthropicanthropic_messages`Reasoning: low, medium, high, xhigh, max128000 token output
Anthropic`anthropic`	`claude-sonnet-4-5`Claude Sonnet · claude-sonnet-4-5	Anthropic`anthropicanthropic_messages`64000 token output
Anthropic`anthropic`	`claude-sonnet-4-5-20250929`Claude Sonnet · claude-sonnet-4-5-20250929	Anthropic`anthropicanthropic_messages`64000 token output
Anthropic`anthropic`	`claude-sonnet-4-6`Claude Sonnet · claude-sonnet-4-6	Anthropic`anthropicanthropic_messages`64000 token output
Anthropic`anthropic`	`claude-sonnet-5`Claude Sonnet · claude-sonnet-5	Anthropic`anthropicanthropic_messages`128000 token output
DeepSeek`deepseek`	`deepseek-reasoner`DeepSeek R1 · deepseek-reasoner	SiliconFlow`siliconflowopenai_chat_completions`Provider-enforced output
DeepSeek`deepseek`	`deepseek-v3.2`DeepSeek V3 · deepseek-v3.2	Baidu Qianfan`baiduopenai_chat_completions`32768 token output
DeepSeek`deepseek`	`deepseek-v4-flash`DeepSeek V4 · deepseek-v4-flash	Baidu Qianfan`baiduopenai_chat_completions`131072 token output DeepSeek`deepseekopenai_chat_completions`Reasoning: none, low, high, maxProvider-enforced output
DeepSeek`deepseek`	`deepseek-v4-pro`DeepSeek V4 · deepseek-v4-pro	Baidu Qianfan`baiduopenai_chat_completions`131072 token output DeepSeek`deepseekopenai_chat_completions`Reasoning: none, low, high, maxProvider-enforced output
ElevenLabs`elevenlabs`	`eleven_english_sts_v2`Eleven Voice Conversion · eleven_english_sts_v2	ElevenLabs`elevenlabselevenlabs_conversion`
ElevenLabs`elevenlabs`	`eleven_flash_v2_5`Eleven Speech · eleven_flash_v2_5	ElevenLabs`elevenlabselevenlabs_speech`
ElevenLabs`elevenlabs`	`eleven_multilingual_sts_v2`Eleven Voice Conversion · eleven_multilingual_sts_v2	ElevenLabs`elevenlabselevenlabs_conversion`
ElevenLabs`elevenlabs`	`eleven_multilingual_v2`Eleven Speech · eleven_multilingual_v2	ElevenLabs`elevenlabselevenlabs_speech`
ElevenLabs`elevenlabs`	`eleven_turbo_v2_5`Eleven Speech · eleven_turbo_v2_5	ElevenLabs`elevenlabselevenlabs_speech`
ElevenLabs`elevenlabs`	`eleven_v3`Eleven Speech · eleven_v3	ElevenLabs`elevenlabselevenlabs_speech`
Baidu`baidu`	`ernie-5.0`ERNIE 5 · ernie-5.0	Baidu Qianfan`baiduopenai_chat_completions`65536 token output
Google`google`	`gemini-3-flash-preview`Gemini · gemini-3-flash-preview	Gemini`geminigemini_interactions`65536 token output
Google`google`	`gemini-3.1-pro-preview`Gemini · gemini-3.1-pro-preview	Vertex AI`vertexvertex_generate_content`Reasoning: low, medium, high65536 token output
Google`google`	`gemini-3.5-flash`Gemini · gemini-3.5-flash	Gemini`geminigemini_interactions`65536 token output Vertex AI`vertexvertex_generate_content`65536 token output
Google`google`	`gemini-3.5-flash-lite`Gemini · gemini-3.5-flash-lite	Vertex AI`vertexvertex_generate_content`Reasoning: minimal, low, medium, high65536 token output
Google`google`	`gemini-3.5-transcribe`Gemini · gemini-3.5-transcribe	Gemini`geminigemini_interactions`
Google`google`	`gemini-3.5-transcribe-preview`Gemini · gemini-3.5-transcribe-preview	Vertex AI`vertexvertex_generate_content`
Google`google`	`gemini-3.6-flash`Gemini · gemini-3.6-flash	Gemini`geminigemini_interactions`Reasoning: minimal, low, medium, high65536 token output
Google`google`	`gemini-3.7-flash`Gemini · gemini-3.7-flash	Gemini`geminigemini_interactions`Reasoning: low, medium, high65536 token output
Google`google`	`gemini-3.8-flash`Gemini · gemini-3.8-flash	Vertex AI`vertexvertex_generate_content`Reasoning: low, medium, high65536 token output
Z.AI`zai`	`glm-5.1`GLM-5 · glm-5.1	Z.AI`zaiopenai_chat_completions`Provider-enforced output
Z.AI`zai`	`glm-5.2`GLM-5 · glm-5.2	Z.AI`zaiopenai_chat_completions`131072 token output
Z.AI`zai`	`glm-asr-2512`GLM ASR · glm-asr-2512	Z.AI`zaimultipart_transcription`
OpenAI`openai`	`gpt-4.1`GPT-4 · gpt-4.1	OpenAI`openaiopenai_responses`Provider-enforced output
OpenAI`openai`	`gpt-4o`GPT-4 · gpt-4o	OpenAI`openaiopenai_responses`Provider-enforced output
OpenAI`openai`	`gpt-4o-mini`GPT-4 · gpt-4o-mini	OpenAI`openaiopenai_responses`Provider-enforced output
OpenAI`openai`	`gpt-5`GPT-5 · gpt-5	OpenAI`openaiopenai_responses`Reasoning: minimal, low, medium, high128000 token output
OpenAI`openai`	`gpt-5-mini`GPT-5 · gpt-5-mini	OpenAI`openaiopenai_responses`Reasoning: minimal, low, medium, highProvider-enforced output
OpenAI`openai`	`gpt-5.5`GPT-5 · gpt-5.5	OpenAI`openaiopenai_responses`Reasoning: none, low, medium, high, xhighProvider-enforced output
OpenAI`openai`	`gpt-5.5-pro`GPT-5 · gpt-5.5-pro	OpenAI`openaiopenai_responses`Reasoning: medium, high, xhighProvider-enforced output
OpenAI`openai`	`gpt-5.6`GPT-5 · gpt-5.6	OpenAI`openaiopenai_responses`Reasoning: none, low, medium, high, xhigh, maxProvider-enforced output
OpenAI`openai`	`gpt-5.6-luna`GPT-5 · gpt-5.6-luna	OpenAI`openaiopenai_responses`Reasoning: none, low, medium, high, xhigh, maxProvider-enforced output
OpenAI`openai`	`gpt-5.6-sol`GPT-5 · gpt-5.6-sol	OpenAI`openaiopenai_responses`Reasoning: none, low, medium, high, xhigh, maxProvider-enforced output
OpenAI`openai`	`gpt-5.6-terra`GPT-5 · gpt-5.6-terra	OpenAI`openaiopenai_responses`Reasoning: none, low, medium, high, xhigh, maxProvider-enforced output
OpenAI`openai`	`gpt-6-astra`GPT-6 · gpt-6-astra	OpenAI`openaiopenai_responses`Reasoning: low, medium, high, xhigh, max128000 token output
OpenAI`openai`	`gpt-image-2`GPT Image · gpt-image-2	OpenAI`openaiopenai_images`
OpenAI`openai`	`gpt-transcribe`GPT Transcribe · gpt-transcribe	OpenAI`openaimultipart_transcription`
xAI`xai`	`grok-4.20-0309-non-reasoning`Grok · grok-4.20-0309-non-reasoning	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-4.20-0309-reasoning`Grok · grok-4.20-0309-reasoning	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-4.3`Grok · grok-4.3	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-4.3-latest`Grok · grok-4.3-latest	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-4.5`Grok · grok-4.5	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-build-0.1`Grok Build · grok-build-0.1	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-code-fast`Grok Code · grok-code-fast	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-code-fast-1`Grok Code · grok-code-fast-1	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-code-fast-1-0825`Grok Code · grok-code-fast-1-0825	xAI`xaixai_responses`Provider-enforced output
xAI`xai`	`grok-imagine-video-1.5`Grok Imagine · grok-imagine-video-1.5	xAI`xaixai_videos_generations`
xAI`xai`	`grok-latest`Grok · grok-latest	xAI`xaixai_responses`Provider-enforced output
Moonshot AI`moonshot`	`kimi-k2.6`Kimi K2 · kimi-k2.6	Moonshot`moonshotopenai_chat_completions`Provider-enforced output
Moonshot AI`moonshot`	`kimi-k2.7-code`Kimi K2 · kimi-k2.7-code	Moonshot`moonshotopenai_chat_completions`Provider-enforced output
Moonshot AI`moonshot`	`kimi-k2.7-code-highspeed`Kimi K2 · kimi-k2.7-code-highspeed	Moonshot`moonshotopenai_chat_completions`Provider-enforced output
Moonshot AI`moonshot`	`kimi-k3`Kimi K3 · kimi-k3	Moonshot`moonshotopenai_chat_completions`Reasoning: low, high, maxProvider-enforced output
MiniMax`minimax`	`minimax-m2`MiniMax M2 · minimax-m2	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.1`MiniMax M2 · minimax-m2.1	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.1-highspeed`MiniMax M2 · minimax-m2.1-highspeed	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.5`MiniMax M2 · minimax-m2.5	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.5-highspeed`MiniMax M2 · minimax-m2.5-highspeed	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.7`MiniMax M2 · minimax-m2.7	MiniMax`minimaxopenai_chat_completions`204800 token output
MiniMax`minimax`	`minimax-m2.7-highspeed`MiniMax M2 · minimax-m2.7-highspeed	MiniMax`minimaxopenai_chat_completions`204800 token output
Meta`meta`	`muse-spark-1.1`Muse Spark · muse-spark-1.1	Meta`metaopenai_chat_completions`Provider-enforced output
Meta`meta`	`muse-spark-1.2`Muse Spark · muse-spark-1.2	Meta`metaopenai_chat_completions`Provider-enforced output
Meta`meta`	`muse-spark-1.3`Muse Spark · muse-spark-1.3	Meta`metaopenai_chat_completions`Reasoning: minimal, low, medium, high, xhigh, maxProvider-enforced output
Alibaba`alibaba`	`qwen-plus`Qwen · qwen-plus	Alibaba Cloud`dashscopedashscope_responses`Provider-enforced output
Marco Polo Research Lab`mprlab`	`qwen3-tts`Qwen3 TTS · qwen3-tts	Dictator`dictatordictator_speech_v1`
Alibaba`alibaba`	`qwen3.6-flash`Qwen · qwen3.6-flash	Alibaba Cloud`dashscopedashscope_responses`65536 token output
Alibaba`alibaba`	`qwen3.7-max`Qwen · qwen3.7-max	Alibaba Cloud`dashscopedashscope_responses`65536 token output
Alibaba`alibaba`	`qwen3.7-plus`Qwen · qwen3.7-plus	Alibaba Cloud`dashscopedashscope_responses`65536 token output
Reve`reve`	`reve-2.1`Reve · 2.1	FAL`falfal_queue_images`
FunAudioLLM`funaudio`	`sensevoice-small`SenseVoice · sensevoice-small	SiliconFlow`siliconflowmultipart_transcription`
Marco Polo Research Lab`mprlab`	`silero-ru`Silero · silero-ru	Dictator`dictatordictator_speech_v1`
OpenAI`openai`	`whisper-base`Whisper · whisper-base	Dictator`dictatordictator_speech_v1`
OpenAI`openai`	`whisper-large-v3`Whisper · whisper-large-v3	Dictator`dictatordictator_speech_v1`
OpenAI`openai`	`whisper-medium`Whisper · whisper-medium	Dictator`dictatordictator_speech_v1`
OpenAI`openai`	`whisper-small`Whisper · whisper-small	Dictator`dictatordictator_speech_v1`
OpenAI`openai`	`whisper-tiny`Whisper · whisper-tiny	Dictator`dictatordictator_speech_v1`
xAI`xai`	`xai-stt`xAI STT · xai-stt	xAI`xaimultipart_transcription`

Provider services

These services use the same provider connection. No model selection is required.

Provider	Service	Input limit	Price
ElevenLabs	Audio alignment	input_audio_bytes: 1000000000 bytes input_audio_seconds: 36000 seconds	0.22 USD/USD/hour
ElevenLabs	Pronunciation dictionary		Exact dictionary creation pricing has not been imported.
HeyGen	Avatar creation	input_image_bytes: 33554432 bytes output_bytes: 268435456 bytes	Exact per-operation pricing depends on account and media duration.
HeyGen	Avatar video generation	input_audio_bytes: 33554432 bytes output_bytes: 268435456 bytes	Exact per-operation pricing depends on account and media duration.
HeyGen	Video lip sync	input_video_bytes: 33554432 bytes input_audio_bytes: 33554432 bytes output_bytes: 268435456 bytes	Exact per-operation pricing depends on account and media duration.
HeyGen	Video translation	input_video_bytes: 33554432 bytes input_audio_bytes: 33554432 bytes output_bytes: 268435456 bytes	Exact per-operation pricing depends on account and media duration.

4 MiBMaximum JSON request body

25 MiBMaximum input audio

3600 secondsMaximum request work budget

Current support contract

The catalog is the source of truth.

The public matrix is generated from the same validated provider registry used by request routing. It states exactly which providers, models, and capabilities the current proxy release supports.

Provider and model routes are explicit, validated runtime contracts.
Unsupported models and capabilities fail before provider dispatch.
The canonical client contract stays stable while provider adapters own upstream API differences.
Provider lifecycle, model-onboarding, and hosted uptime commitments are outside the current catalog contract pending an approved support and SLA policy.

One integration, ongoing choice

Start once. Keep choosing the model that fits.

Choose HTTP, Go, Python, or the command line, then use the current model catalog to select the route your product needs.

Choose an integration Read implementation guides