OpenRouter

Access and compare models through one API while simplifying integrations, routing around provider issues, and tracking usage and costs.

Compare responses from hundreds of AI models on the same prompt in one interface to evaluate options efficiently.

Keep inference in the EU or US while giving teams expanded workspace capacity, identity options, spend controls, and operational alerts.

Gives organizations governed access to hundreds of AI models through one gateway, with resilient routing, spend oversight, security controls, and consolidated enterprise procurement.

Build with OpenRouter models using typed clients that reduce boilerplate and provide consistent capabilities across TypeScript, Python, and Go.

Build agents with less orchestration by automating multi-turn tool execution and state while routing model calls across providers.

Manage OpenRouter configuration as code to make platform changes repeatable and reviewable.

Review and replay SDK activity locally to inspect requests, token counts, and errors during development.

Connect AI assistants to OpenRouter models and current platform information without installing a server or relying on stale reference data.

Run familiar agent CLIs with OpenRouter models, existing organization policies, and consolidated usage tracking.

Work with multiple OpenRouter models from a native terminal coding agent.

Find and compare models using evaluations based on your project’s own prompts, with results that help you weigh performance, time, and cost.

Help teams delegate work in Slack using model-flexible interns that learn reusable skills and can improve across the organization.

Send tasks to OpenRouter models from your Mac and manage their sessions and approvals in one desktop experience.

Access aggregated daily model usage rankings as JSON to work with OpenRouter’s public rankings data programmatically.

Infrastructure for routing application requests across LLM providers through a unified interface, supporting provider portability and structured generation.

Products: OpenRouter API, OpenRouter Business, OpenRouter Enterprise, OpenRouter Client SDKs, OpenRouter MCP Server, Ori Harness

Runtime controls that let enterprises enforce access permissions, isolate provider credentials, and apply hierarchical usage policies before AI requests reach configured providers.

Products: OpenRouter Business, OpenRouter Enterprise, OpenRouter API, Ori Harness

Comparison of model responses, task performance, latency, and costs to support model selection, using side-by-side testing or project-specific evaluations.

Products: Ori Eval, OpenRouter Chat

Agent-based assistance for software development through terminal workflows, with model selection across multiple providers.

Products: Ori Code

OpenRouter API

OpenRouter API provides a unified, OpenAI-compatible interface for accessing hundreds of AI models across multiple providers. Developers can use one API key and consistent request formats to switch models without changing their integration. OpenRouter can route requests based on price, throughput, latency, and reliability, and automatically retry through other providers when failures or rate limits occur. The API supports streaming, tool calling, structured outputs, and models for text, images, video, and audio. Developers can also configure provider preferences, data policies, and regional routing, and track usage and costs across their requests.

• Access Hundreds Of Models
  Use one API key and endpoint to call models from multiple providers, making it possible to compare or change models without building a separate integration for each provider.
• Route Requests To Suitable Providers
  Let OpenRouter select providers using measured performance and price factors, or set provider preferences to prioritize attributes such as lower price, higher throughput, or lower latency.
• Reuse Familiar API Integrations
  Connect using a familiar request format or point an existing OpenAI SDK integration at OpenRouter, reducing the changes needed to access its model catalog.
• Retry Through Available Providers
  Keep requests moving when a provider is unavailable or rate-limited by retrying through other eligible providers or models; failed attempts are not charged for model tokens.
• Apply Data And Provider Policies
  Set restrictions on eligible providers, models, and data regions, including routing requests only to zero-data-retention endpoints where available.
• Build Across Multiple Modalities
  Use OpenRouter to access models for a range of tasks, including text generation, image and video generation, speech transcription, and speech synthesis.
• Stream Responses And Call Tools
  Deliver responses as they are generated and use model tool calls in applications, with routing that accounts for provider support where possible.
• Keep Inference Within A Region
  Send requests to a regional API endpoint to have eligible shared provider endpoints process prompts and completions in that region; requests fail rather than silently routing out of region.
• Track Usage And Costs
  Review request activity and spending across models and providers, and include usage information in responses to support application-level cost tracking.
• Request Structured JSON Responses
  Specify JSON object or JSON schema response formats to request structured output from models that support the capability.
• Extend Model Capabilities
  Enable supported plugins in API requests to add capabilities such as web search, PDF processing, automatic JSON repair, or prompt compression.

OpenRouter Chat

OpenRouter Chat is a playground for testing and evaluating responses from OpenAI, Google, Anthropic, and more than 600 AI models in one chat interface. Choose one or more models, send the same message to each, and read their responses side by side. Responses are AI-generated and may be inaccurate, so review them before relying on them.

• Compare Responses Side By Side
  Submit the same prompt to multiple selected models and view their answers together, making it easier to evaluate how models respond to a given request.
• Access Hundreds Of AI Models
  Explore and test models from multiple providers in one place, without switching between separate chat interfaces.

OpenRouter Business

OpenRouter Business is a self-serve plan for routing inference requests to providers in the EU or US, using the same API key and account. It also includes expanded workspaces, advanced notifications, workload identity federation, and guardrails and spend controls. The plan costs 8% on credit purchases, with no monthly minimum or contract. Its DPA applies through the Terms of Service, and organizations can request a signed copy.

• Keep Inference Within Your Chosen Region
  Send requests to the EU or US regional endpoint to have eligible inference processed in that region. If no eligible in-region provider can serve a request, it fails instead of being routed elsewhere.
• Enforce Allowed Request Regions
  Set permitted data regions in guardrails so requests arriving through an unapproved regional domain are rejected before processing.
• Separate Up To 1,000 Workspaces
  Organize teams, projects, and environments in separate workspaces, with a Business plan limit of 1,000.
• Authenticate Services With Identity Provider Tokens
  Let services authenticate using tokens from your identity provider rather than long-lived API keys.
• Apply Guardrails And Manage Spend
  Use the plan's guardrails and spend controls to manage organizational policies and spending across OpenRouter usage.
• Send Operational Alerts To Your Team
  Use email for workspace budgets, API key spend limits, and new model alerts. Send workspace budget and model alerts to Slack or a custom webhook as well.
• Pay Only An 8% Credit-Purchase Fee
  Pay an 8% platform fee when purchasing credits while inference is billed at provider rates. Business has no monthly minimum or contract.
• Document Your Data Processing Terms
  Rely on the DPA that applies through the Terms of Service, or request a signed copy for your organization.

OpenRouter Enterprise

OpenRouter Enterprise is an AI gateway and inference control plane that gives organizations access to 600+ models across 80+ providers through one API. It offers model and provider routing, automatic failover, spend controls, security and data policies, and enterprise identity and reporting features. Enterprise agreements include invoiced billing options, credit lines, an SLA, priority support, and a dedicated engineering contact.

• Access Hundreds Of Models Through One API
  Teams can use one API and key for models and providers they approve, and switch models without changing their integration.
• Enforce Data Policies On Every Request
  Teams can apply retention, logging, and guardrail requirements to requests and restrict routing to providers and models that meet their policies.
• Match Models To Each Task
  Teams can route prompts to models suited to different tasks and access new model releases without making code changes.
• Route Requests By Price And Performance
  Teams can use provider routing informed by live latency and error rates, and favor lower-cost providers for the same model.
• Retry Requests Across Providers And Models
  Requests can be retried on another provider or model during outages or rate limits, while failed or empty completions are not charged.
• Set Budgets And Track Spend
  Teams can set spending limits at multiple organizational levels, receive alerts, and review costs by team, model, and user.
• Keep Inference Within Your Chosen Region
  Teams can route inference through EU or US endpoints and avoid out-of-region fallback; requests fail when no eligible endpoint in the selected region can serve them.
• Manage Team And Service Access
  Organizations can connect team sign-in and access provisioning to their identity provider, and authenticate services without long-lived API keys.
• Consolidate Provider Billing And Contracts
  Organizations can manage providers through one agreement and invoice, choose invoiced billing or card payments, and use credit lines to keep running.
• Use Your Provider Keys And Capacity
  Organizations can retain their provider-negotiated rates and capacity while using OpenRouter for routing, failover, budgets, and logging.
• Review And Export Organization-Wide Usage
  Teams can review and export requests across users and examine spending across organizational teams, models, and users.
• Get Priority Support And An SLA
  Enterprise customers receive contractual service-level agreements, priority support channels, and a dedicated engineering contact.
• Support Security And Legal Reviews
  Organizations can access compliance documentation and request OpenRouter's DPA or a review of their own agreement as part of security and legal reviews.

OpenRouter Client SDKs

Typed client libraries for TypeScript, Python, and Go that wrap the OpenRouter API. The SDKs provide full type safety, types generated from the OpenAPI specification, and a lean layer over the REST API. They support streaming, tool calls, structured output, and embeddings, with consistent behavior across languages. Use them to access OpenRouter models through a single client integration.

• Choose Your Language
  Integrate with OpenRouter from TypeScript, Python, or Go using client libraries designed to provide a consistent development experience.
• Write With Typed Clients
  Use type-safe client interfaces and generated API types when making OpenRouter requests, with minimal overhead over the REST API.
• Stream Across Languages
  Use streaming in the supported language clients with consistent behavior across TypeScript, Python, and Go.
• Handle Tool Calls Consistently
  Work with tool calls through the client SDKs across supported languages. The Client SDK returns calls for your application to dispatch.
• Use Structured Output
  Access structured output through the client SDKs with consistent behavior across TypeScript, Python, and Go.
• Access Embeddings
  Use the Client SDK documentation to integrate embeddings alongside the other documented API capabilities.
• Configure A Regional Endpoint
  Set the SDK server URL to a regional OpenRouter endpoint to send requests through that region. Availability and routing depend on the account plan and eligible providers.

OpenRouter Agent SDK

The OpenRouter Agent SDK (@openrouter/agent) provides higher-level primitives for building AI agents that use OpenRouter models. Its callModel function manages multi-turn conversation loops, tool execution, and state, while supporting type-safe tools, streaming, and stop conditions. Agents can route calls across providers through OpenRouter. The SDK is available for TypeScript and Python, with Go in beta.

• Automate Multi-Turn Agent Loops
  Build agents that can continue conversations through multiple model and tool interactions without manually coordinating each turn.
• Run Tools Within Agent Loops
  Connect model responses to application functions and have the SDK manage tool execution and result handoff during an agent interaction.
• Route Agent Calls Across Providers
  Build agents that call models through OpenRouter and route across providers rather than being limited to a single provider.
• Define Tools With Typed Inputs
  Specify tool inputs through schemas so tools can be defined with structured, typed parameters.
• Manage Agent State Automatically
  Rely on the SDK to manage agent state as part of the callModel workflow.
• Stream Agent Responses
  Use streaming in agent workflows to receive response output progressively rather than only after generation is complete.
• Set Agent Stop Conditions
  Specify conditions, such as a maximum number of steps, to determine when an agent should stop its workflow.
• Build Agents In Multiple Languages
  Use the Agent SDK with TypeScript or Python, or try its beta Go support.

OpenRouter Terraform Provider

The OpenRouter Terraform Provider lets teams manage OpenRouter as infrastructure-as-code, with repeatable and reviewable configuration for their platform.

• Manage OpenRouter As Code
  Define and manage OpenRouter configuration as code, making platform configuration repeatable and reviewable.

OpenRouter DevTools

OpenRouter DevTools captures SDK telemetry during development and lets you review and replay requests in a local web viewer, including their token counts and errors.

• Capture SDK Telemetry
  Collect SDK telemetry during development for review in the local web viewer.
• Inspect Requests Locally
  Review captured requests, token counts, and errors in a local web viewer.
• Replay Captured Requests
  Replay requests during development from the local web viewer.

OpenRouter MCP Server

OpenRouter’s remote-hosted Model Context Protocol server connects compatible clients, including Claude Code and Cursor, to OpenRouter’s catalog of more than 600 models. Coding assistants can also retrieve live information about available models, pricing, credit balance, and usage rankings, and search OpenRouter documentation. Connect by adding the server URL to an MCP client and approving an OAuth login; there is nothing to install.

• Access 600+ Models
  Give compatible MCP clients access to OpenRouter’s catalog of more than 600 models.
• Check Available Models
  Help assistants use current model availability information rather than relying on stale training knowledge.
• Connect Without Installing
  Use the remotely hosted server by adding one URL to an MCP client and approving an OAuth login, without installing the server.
• Check Model Prices
  Let assistants consult current model pricing when helping users build with OpenRouter.
• Search OpenRouter Docs
  Let assistants find relevant OpenRouter documentation during development instead of relying solely on previously learned information.
• Check Credit Balance
  Make account credit balance available to assistants through the MCP connection.
• View Usage Rankings
  Enable assistants to consult current usage rankings while working with OpenRouter information.

Ori Harness

Ori Harness is a command-line tool that runs supported agent CLIs through OpenRouter, giving them access to OpenRouter models and organization settings. Launch agents such as Claude Code, Codex, Grok Build, Hermes, OpenCode, Pi, Cline, and others with the ori command. Choose an OpenRouter model while passing the rest of your command and flags through to the agent unchanged. Sign in with OAuth using an existing OpenRouter account, apply organization guardrails to requests, and view usage and spend from agents in the OpenRouter dashboard.

• Run Familiar Agent CLIs
  Use supported agent CLIs through OpenRouter while continuing to work with the agents and commands you already know.
• Choose From OpenRouter Models
  Choose an OpenRouter model for supported agents without replacing the agent CLI, and pass the rest of your command through to the agent.
• Enforce Organization Policies
  Apply organization model allowlists, budgets, and workspace permissions across supported agents, with out-of-policy requests blocked before they run.
• Sign In With Your OpenRouter Account
  Authenticate through an existing OpenRouter login instead of creating or managing a separate API key for Ori.
• Track Agent Spend In One Place
  View usage and spend across supported agents in the OpenRouter dashboard, with agent-level cost breakdowns and one combined bill.
• Install Missing Agents With Guidance
  Get installation instructions and an offer to install an agent when Ori finds that its CLI is not already available on your PATH.

Ori Code

Ori Code is OpenRouter’s native coding agent for the terminal, built from the ground up for working with multiple models through OpenRouter. It provides a terminal experience crafted specifically for that workflow.

• Work With Models In The Terminal
  Use Ori Code as a terminal-native coding agent designed for working with multiple models through OpenRouter.
• Choose From Multiple OpenRouter Models
  Work with multiple models from OpenRouter in Ori Code’s terminal coding environment.

Ori Eval

Ori Eval helps teams identify the best model for their project by using an agent to find relevant test material in their code, write evaluations, and run selected models against them. It ranks candidates using evaluation results, with scores, run times, and costs available for comparison. Evaluations run on a pinned harness and model, and can use LLM-as-judge grading.

• Rank Models By Evaluation Results
  Review ranked model candidates and their scores, run times, and costs to make a project-specific model choice.
• Find Relevant Project Prompts
  Use prompts and test material from your application to evaluate models against work that reflects your project.
• Test Models On Your Prompts
  Compare candidate models by running them against evaluation prompts drawn from your own project.
• Create Project-Based Evaluations
  Turn project-specific test material into evaluations, including evaluations represented as *.eval.ts files.
• Keep Evaluation Runs Consistent
  Use a pinned harness and model for evaluation authoring and grading, providing a consistent setup for comparing candidate models.
• Grade Model Outputs
  Use LLM-as-judge grading to assess outputs produced during model evaluations.

Ori Intern

Ori Intern lets teams deploy agents in Slack that can use any model. Interns learn and build skills naturally, and those skills can be shared across interns so teams can improve together. Skills work across models, helping teams delegate work using a cost-optimal model. Ori Intern is in private testing and available through a waitlist.

• Delegate Work From Slack
  Give team members an Intern in Slack and assign work through messages, such as requesting a weekly team update.
• Use Skills Across Models
  Use an Intern with any model while keeping its skills usable across models.
• Build Skills Through Work
  Let Interns learn and build skills naturally as they perform delegated work.
• Share Skills Across Interns
  Make skills available to other Interns so improvements and learned capabilities can be shared across the organization.
• Choose Cost-Optimal Models
  Select a cost-optimal model for delegated work without limiting skills to a single model.

Ori Desktop

Ori Desktop is an upcoming, OS-native Mac experience for sending tasks to OpenRouter models. Tasks run with Ori and can use any harness across hundreds of models. Users can create tasks from the menu bar, manage sessions, and make approvals. The product is currently available by waitlist.

• Run Tasks From Your Mac
  Use a native desktop experience to send tasks to OpenRouter models from your Mac.
• Choose Models And Harnesses
  Run tasks with Ori using any harness and models available through OpenRouter.
• Create Tasks From The Menu Bar
  Start tasks directly from the menu bar on your Mac.
• Manage Task Sessions
  Keep task sessions manageable through the Ori Desktop experience.
• Make Task Approvals
  Handle approvals within the desktop task workflow.

OpenRouter Data API

The OpenRouter Data API provides JSON access to the same aggregated usage data that powers OpenRouter’s public rankings. It returns the top 50 models per day by total token usage and authenticates with an existing OpenRouter API key. OpenRouter describes shared data as aggregated usage trends and says it does not sell prompt data.

• Retrieve Daily Model Rankings
  Access the top 50 models for each day by total token usage through an API, using the same data that powers the public rankings page.
• Access Rankings Data As JSON
  Retrieve rankings data in a structured JSON format for programmatic access.
• Authenticate With Your Existing API Key
  Use an existing OpenRouter API key to authenticate API requests.
• Use Aggregated Usage Trends
  Work with aggregated usage trends. OpenRouter states that it does not sell prompt data.