---
title: "Supported API Endpoints: Chat, Responses, Embeddings, Audio"
url: "/docs/supported_endpoints"
canonical_url: "https://docs.litellm.ai/docs/supported_endpoints"
type: "docs"
last_updated: "2026-11-19"
summary: "LiteLLM serves 40+ API endpoints in OpenAI, Anthropic, and Google formats from the Python SDK and the AI Gateway. Find an endpoint and open its guide."
related:
  - "/docs/self_hosted_coding_agents/moyai/architecture"
  - "/docs/assistants"
---
# Supported API Endpoints: Chat, Responses, Embeddings, Audio

> Index of all LiteLLM docs: https://docs.litellm.ai/llms.txt


LiteLLM serves chat, responses, embeddings, images, video, audio, realtime, rerank, vector store, file, batch, MCP, and A2A endpoints. Send a request in the OpenAI, Anthropic, or Google format, and LiteLLM calls any provider that supports it. Use the Python SDK or the AI Gateway. Most endpoints work in both.

## Which endpoints does LiteLLM support? {#all-endpoints}

Search by name, path, or Python function, or select a category. Each card opens the guide for that endpoint.

### Chat and responses

Send prompts to any model in the OpenAI, Anthropic, Google or Bedrock format.

- [Chat Completions](https://docs.litellm.ai/docs/completion/input.md) (`POST /v1/chat/completions`): Call any supported model in the OpenAI chat format, with streaming, tools and structured output.
- [Text Completions](https://docs.litellm.ai/docs/text_completion.md) (`POST /v1/completions`): Legacy OpenAI text completion format, bridged to chat models.
- [Responses](https://docs.litellm.ai/docs/response_api.md) (`POST · GET /v1/responses`): OpenAI Responses API for any model, with stateful sessions, tools and background mode.
- [Anthropic Messages](https://docs.litellm.ai/docs/anthropic_unified/index.md) (`POST /v1/messages`): Call any model in the Anthropic Messages format, so Claude Code and the Anthropic SDK work unchanged.
- [Google generateContent](https://docs.litellm.ai/docs/generateContent.md) (`POST /v1beta/models/{model}:generateContent`): Call any model in the Google generateContent format, with streaming through streamGenerateContent.
- [Interactions](https://docs.litellm.ai/docs/interactions.md) (`POST · GET /v1beta/interactions`): Google Interactions API for any model, with stateful conversations and streaming.
- [Bedrock Converse](https://docs.litellm.ai/docs/bedrock_converse.md) (`POST /bedrock/model/{model}/converse`): Native Bedrock Converse calls through the proxy with LiteLLM keys, budgets and logging.
- [Bedrock Invoke](https://docs.litellm.ai/docs/bedrock_invoke.md) (`POST /bedrock/model/{model}/invoke`): Native Bedrock InvokeModel calls through the proxy with LiteLLM keys, budgets and logging.

### Embeddings, rerank and retrieval

Embed text, rank results, search the web and query vector stores.

- [Embeddings](https://docs.litellm.ai/docs/embedding/supported_embedding.md) (`POST /v1/embeddings`): Turn text into vectors with any supported embedding model in the OpenAI format.
- [Rerank](https://docs.litellm.ai/docs/rerank.md) (`POST /v1/rerank`): Rank documents by relevance to a query, in the Cohere rerank format.
- [Search](https://docs.litellm.ai/docs/search/index.md) (`POST /v1/search`): One web search API over Perplexity, Tavily, Exa, Brave and other search providers.
- [Vector Stores](https://docs.litellm.ai/docs/vector_stores/index.md) (`POST · GET /v1/vector_stores`): Create, register and search vector stores on OpenAI, Bedrock, Gemini and more.
- [Vector Store Files](https://docs.litellm.ai/docs/vector_store_files.md) (`POST · GET /v1/vector_stores/{vector_store_id}/files`): Add, list, read and delete the files inside a vector store.
- [RAG Ingest](https://docs.litellm.ai/docs/rag_ingest.md) (`POST /v1/rag/ingest`): Upload, chunk, embed and store documents in a vector store with one call.
- [RAG Query](https://docs.litellm.ai/docs/rag_query.md) (`POST /v1/rag/query`): Search a vector store, optionally rerank, and answer with an LLM in one call.

### Audio and realtime

Transcribe speech, generate speech and stream live audio.

- [Audio Transcriptions](https://docs.litellm.ai/docs/audio_transcription.md) (`POST /v1/audio/transcriptions`): Convert speech to text with Whisper, Deepgram, Gemini and other speech models.
- [Audio Speech](https://docs.litellm.ai/docs/text_to_speech.md) (`POST /v1/audio/speech`): Convert text to speech with OpenAI, Azure, AWS Polly, ElevenLabs and other providers.
- [Realtime](https://docs.litellm.ai/docs/realtime.md) (`WS /v1/realtime`): Stream live audio and text over a WebSocket, load balanced across OpenAI, Azure and more.
- [Realtime Translations](https://docs.litellm.ai/docs/realtime_translation.md) (`WS /v1/realtime/translations`): Live speech-to-speech translation with gpt-realtime-translate over WebSocket or WebRTC.

### Images and video

Generate and edit images and video.

- [Image Generations](https://docs.litellm.ai/docs/image_generation.md) (`POST /v1/images/generations`): Generate images from a text prompt with DALL-E, Imagen, Stable Diffusion and others.
- [Image Edits](https://docs.litellm.ai/docs/image_edits.md) (`POST /v1/images/edits`): Edit one or more images with a text prompt, in the OpenAI images edits format.
- [Image Variations](https://docs.litellm.ai/docs/image_variations.md) (`SDK litellm.image_variation()`): Create variations of an image with OpenAI or Topaz. Python SDK only, with no Gateway route.
- [Videos](https://docs.litellm.ai/docs/videos.md) (`POST · GET /v1/videos`): Generate, remix and download videos with Sora, Veo and other video models.

### Documents

Read text and typed fields from documents and images.

- [OCR](https://docs.litellm.ai/docs/ocr.md) (`POST /v1/ocr`): Extract text from PDFs and images with Mistral, Azure AI, Gemini Enterprise Agent Platform or Cohere.
- [Extraction](https://docs.litellm.ai/docs/extraction.md) (`POST /v1/extraction`): Fill typed fields from text and images, with confidence scores. Runs on Cloudflare Clairic.

### Agents and tools

Run agents, call MCP tools and manage code containers and memory.

- [Agent Gateway (A2A)](https://docs.litellm.ai/docs/a2a.md) (`POST /a2a/{agent_id}`): Invoke agents in the A2A protocol and control which keys and teams can reach each agent.
- [MCP Gateway](https://docs.litellm.ai/docs/mcp.md) (`POST /mcp`): One MCP endpoint for all your tool servers, with access control by key and team.
- [Assistants](https://docs.litellm.ai/docs/assistants.md) (`POST · GET /v1/assistants`): OpenAI Assistants, threads and runs. OpenAI shut this API down on August 26, 2026. Use Responses.
- [Managed Agents](https://docs.litellm.ai/docs/managed_agents.md) (`POST · GET /v1beta/agents`): Create and run Google AI Studio managed agents through the proxy.
- [Skills](https://docs.litellm.ai/docs/skills.md) (`POST · GET /v1/skills`): Create, list and use Anthropic skills as reusable agent capabilities.
- [Containers](https://docs.litellm.ai/docs/containers.md) (`POST · GET /v1/containers`): Create and manage OpenAI and Azure code interpreter containers.
- [Container Files](https://docs.litellm.ai/docs/container_files.md) (`POST · GET /v1/containers/{container_id}/files`): Upload, list and download files in a code interpreter container.
- [Sandbox](https://docs.litellm.ai/docs/sandbox.md) (`POST /v1/responses`): Run model-written code in e2b or OpenSandbox with the code_interpreter tool on /v1/responses.
- [Memory](https://docs.litellm.ai/docs/memory_management.md) (`POST · GET /v1/memory`): Store and read key-value memory entries scoped to users and teams.

### Safety and routing

Moderate content, run guardrails and get model decisions.

- [Moderations](https://docs.litellm.ai/docs/moderation.md) (`POST /v1/moderations`): Check text for harmful content in the OpenAI moderations format. No model setup is needed.
- [Apply Guardrail](https://docs.litellm.ai/docs/apply_guardrail.md) (`POST /guardrails/apply_guardrail`): Call a guardrail set up on your proxy directly from any service.
- [Decisions](https://docs.litellm.ai/docs/decisions.md) (`POST /v1/decisions`): Ask a typed question about an input and get probabilities back instead of text.

### Files, batches and tuning

Upload files, run batch jobs, fine tune models and run evals.

- [Files](https://docs.litellm.ai/docs/files_endpoints.md) (`POST · GET /v1/files`): Upload and manage files on the provider, in the OpenAI files format.
- [Batches](https://docs.litellm.ai/docs/batches.md) (`POST · GET /v1/batches`): Run large async jobs at lower cost on OpenAI, Azure, Bedrock, Gemini Enterprise Agent Platform and more.
- [Fine Tuning](https://docs.litellm.ai/docs/fine_tuning.md) (`POST · GET /v1/fine_tuning/jobs`): Create and track fine tuning jobs on OpenAI, Azure OpenAI and Gemini Enterprise Agent Platform.
- [Evals](https://docs.litellm.ai/docs/evals_api.md) (`POST · GET /v1/evals`): Create and run OpenAI evals to measure model quality against test criteria.

### Tokens

Count and inspect tokens before you send a request.

- [Token Counting](https://docs.litellm.ai/docs/count_tokens.md) (`POST /v1/responses/input_tokens`): Count input tokens with the provider token counting API before you send a request.
- [Anthropic Count Tokens](https://docs.litellm.ai/docs/anthropic_count_tokens.md) (`POST /v1/messages/count_tokens`): Count tokens for a messages request in the Anthropic format.
- [Tokenize](https://docs.litellm.ai/docs/tokenize.md) (`POST /tokenize`): Get the token count, IDs and text for one string from vLLM or IBM watsonx.ai. Not in a release yet.

## How does LiteLLM handle a request? {#how-it-works}

LiteLLM handles a request in one of three ways. The diagram shows each path from your app to the provider.

- Translated: your app sends a chat request in the OpenAI, Anthropic, or Google format. LiteLLM changes it to the API of the provider you select and sends the answer back in the format your app sent.
- Native: your app sends one request shape for each endpoint (for example `/v1/embeddings`). LiteLLM maps it to the API of each provider that supports it.
- Pass-through: your app calls the provider API through LiteLLM (for example `/anthropic/v1/messages`). LiteLLM checks the LiteLLM key, adds the provider credentials, and sends the request without changes. Cost tracking depends on the provider.

### Compare the three paths {#request-paths}

| | Translated | Native | Pass-through |
| --- | --- | --- | --- |
| **What LiteLLM does** | Takes a chat request in one of three formats and maps it to the API of any chat provider | Takes one request shape for each endpoint and maps it to the API of each provider that supports it | Sends your request to the provider API without changes |
| **Request format** | OpenAI, Anthropic, or Google | Usually OpenAI. Rerank uses the Cohere format, search the Perplexity format, and OCR the Mistral format | The API of the provider |
| **Examples** | `/v1/chat/completions`, `/v1/responses`, `/v1/messages`, `generateContent` | `/v1/embeddings`, `/v1/images/generations`, `/v1/audio/speech`, `/v1/rerank` | `/anthropic/v1/messages`, `/bedrock/...`, `/gemini/...` |
| **LiteLLM keys and logs** | Yes | Yes | Yes |
| **Cost tracking** | Yes | Yes | For the providers that support it. Each provider page says if it does |
| **Use it when** | You want one client for many models | You use one capability across providers | You already use the provider SDK, or you need a provider feature that LiteLLM does not map |

## Which providers have pass-through endpoints? {#pass-through-endpoints}

A pass-through route sends a request to the API of one provider. LiteLLM checks your LiteLLM key, adds the provider credentials, and logs the request. Cost tracking depends on the provider. Read [why to use pass-through endpoints](./pass_through/intro.md), or select a provider.

- [Anthropic](https://docs.litellm.ai/docs/pass_through/anthropic_completion.md) (`/anthropic`)
- [AssemblyAI](https://docs.litellm.ai/docs/pass_through/assembly_ai.md) (`/assemblyai`)
- [AWS Bedrock](https://docs.litellm.ai/docs/pass_through/bedrock.md) (`/bedrock`)
- [Azure OpenAI](https://docs.litellm.ai/docs/pass_through/azure_passthrough.md) (`/azure`)
- [Azure Speech](https://docs.litellm.ai/docs/pass_through/azure_speech.md) (`/azure_speech`)
- [Cohere](https://docs.litellm.ai/docs/pass_through/cohere.md) (`/cohere`)
- [Comprehend Medical](https://docs.litellm.ai/docs/pass_through/comprehend_medical.md) (`/comprehendmedical`)
- [Cursor](https://docs.litellm.ai/docs/pass_through/cursor.md) (`/cursor`)
- [Deepgram live transcription](https://docs.litellm.ai/docs/pass_through/deepgram_listen_websocket.md) (`/deepgram/v1/listen`)
- [EU JEV](https://docs.litellm.ai/docs/pass_through/eu_jev.md) (`/eu_jev/v1/systemone`)
- [GigaChat](https://docs.litellm.ai/docs/pass_through/gigachat.md) (`/gigachat`)
- [Google AI Studio](https://docs.litellm.ai/docs/pass_through/google_ai_studio.md) (`/gemini`)
- [Langfuse](https://docs.litellm.ai/docs/pass_through/langfuse.md) (`/langfuse`)
- [Mistral](https://docs.litellm.ai/docs/pass_through/mistral.md) (`/mistral`)
- [OpenAI](https://docs.litellm.ai/docs/pass_through/openai_passthrough.md) (`/openai_passthrough`)
- [TinyFish](https://docs.litellm.ai/docs/pass_through/tinyfish.md) (`/tinyfish`)
- [Amazon Transcribe](https://docs.litellm.ai/docs/pass_through/transcribe.md) (`/transcribe`)
- [Typesafe](https://docs.litellm.ai/docs/pass_through/typesafe.md) (`/typesafe`)
- [Managed IDs](https://docs.litellm.ai/docs/proxy/passthrough_managed_ids.md)
- [Gemini Enterprise Agent Platform (formerly Vertex AI)](https://docs.litellm.ai/docs/pass_through/vertex_ai.md) (`/vertex_ai`)
- [Gemini Enterprise Agent Platform Live API](https://docs.litellm.ai/docs/pass_through/vertex_ai_live_websocket.md) (`/vertex_ai/live`)
- [Gemini Enterprise Agent Platform Search](https://docs.litellm.ai/docs/pass_through/vertex_ai_search_datastores.md) (`/vertex_ai/discovery`)
- [vLLM](https://docs.litellm.ai/docs/pass_through/vllm.md) (`/vllm`)
- [Custom pass-through routes](https://docs.litellm.ai/docs/proxy/pass_through.md)
- [Cost tracking](https://docs.litellm.ai/docs/proxy/pass_through_cost_tracking.md)
- [Guardrails](https://docs.litellm.ai/docs/proxy/pass_through_guardrails.md)

## Frequently asked questions {#faq}

### Which endpoints does LiteLLM support?

LiteLLM supports more than 40 endpoints. They include chat completions, responses, Anthropic messages, embeddings, images, video, audio, realtime, rerank, vector stores, files, batches, MCP, and A2A. This page lists them, with a link to the guide for each.

### Can I send an Anthropic-format request to a model that is not from Anthropic?

Yes. Send the request to `/v1/messages` in the Anthropic format and set the model to any chat model in your config. LiteLLM changes the request to the API of that provider and sends the answer back in the Anthropic format. See [/v1/messages](./anthropic_unified/index.md).

### Can I call a model that only supports the Responses API from /chat/completions?

Yes. LiteLLM bridges `/chat/completions` to the Responses API when the provider does not support chat completions for that model. The reverse also works. `/v1/responses` can call providers that only have chat completions. See [/responses](./response_api.md).

### What is the difference between a translated endpoint and a pass-through endpoint?

A translated endpoint takes one request format and calls many providers. A pass-through endpoint sends your request to the API of one provider without changes. Both use your LiteLLM key and are logged. For pass-through endpoints, cost tracking depends on the provider.

### Do the endpoints work in the Python SDK and in the AI Gateway?

Most endpoints have a Python SDK function and a Gateway route. Each card on this page shows the Python function when one exists. Serving MCP servers is a Gateway feature only.

### How do I ask for an endpoint that is not on this page?

[Open an issue](https://github.com/BerriAI/litellm/issues/new) in the BerriAI/litellm repository on GitHub with the endpoint and the providers you need.

## Next steps {#next-steps}

- **Tutorial:** [Start the AI Gateway](./proxy/docker_quick_start.md) or [make your first SDK call](./learn/sdk_quickstart.md).
- **How-to:** [Call the Gateway from the OpenAI SDK, LangChain, or curl](./proxy/user_keys.md).
- **Explanation:** [Why use pass-through endpoints](./pass_through/intro.md).
- **Reference:** [All model providers](./providers/index.mdx) and the endpoints each one supports.

## Related pages

- [Architecture](https://docs.litellm.ai/docs/self_hosted_coding_agents/moyai/architecture.md)
- [/assistants](https://docs.litellm.ai/docs/assistants.md)
