OpenRouter
by OpenRouter
One OpenAI-compatible API that routes every request across 500+ models and 80+ inference providers
OpenRouter is a unified LLM gateway that gives engineering teams a single OpenAI-compatible endpoint to more than 500 models from over 80 inference providers, with automatic failover, per-request routing on price, latency and reliability, unified billing and org-level data policies. It is for platform and AI teams who want multi-model access without negotiating, integrating and monitoring each provider separately.
OpenRouter is an AI model gateway founded in 2023 by OpenSea co-founder Alex Atallah and Louis Vichy. It exposes a single OpenAI-compatible API that fronts more than 500 text, image, video and audio models from over 80 inference providers, including Anthropic, Google, OpenAI, xAI and DeepSeek, so an application can change models with a string rather than a re-integration. Each request is evaluated and routed to a provider based on task requirements, price, speed and reliability, and the platform automatically falls back to another provider when one is degraded or down — the availability argument that distinguishes it from calling a single vendor directly. OpenRouter passes provider inference pricing through without a markup and monetises through a 5.5% fee on credit purchases and a 5% fee on bring-your-own-key usage above a monthly allowance. Enterprise controls include SOC 2 compliance, GDPR-compatible infrastructure, zero-data-retention provider routing, zero prompt/completion logging by default, SSO, organisation management, per-key credit limits, programmatic key management and provider/region exclusion for data-residency requirements. Volume reached roughly 25 trillion tokens per week by May 2026, a 5x increase in six months, and the company reports 10M+ users and 250k+ applications. Named customers include NVIDIA, Zoom, AMD, Replit, Framer, Webflow and Lovable. OpenRouter raised a $40M seed and Series A in June 2025 co-led by a16z and Menlo Ventures, then a $113M CapitalG-led Series B in May 2026 at roughly a $1.3B valuation. On 19 August 2026 Stripe announced an agreement to acquire the company, with terms undisclosed.
The platform or AI engineering lead who has to support several frontier and open-weight models across products and does not want a separate contract, SDK, quota and billing relationship with each provider.
Swap or fail over between 500+ models behind one OpenAI-compatible endpoint and one invoice, without rewriting application code or re-negotiating provider terms.
At a Glance
- Category
- AI Models & APIs
- Pricing
- Usage-based, Freemium, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, Platform Engineers, AI Engineers
- Deployment
- API-based, Cloud-only
- Founded
- 2023
- Customers
- 10M+ users and 250k+ applications (vendor-reported); named enterprise customers include NVIDIA, Zoom, AMD, Replit, Framer, Webflow and Lovable
Key Features
- ✓Unified OpenAI-compatible API
One endpoint and request schema for 500+ models across 80+ providers, so switching models is a config change rather than a re-integration.
- ✓Per-request model routing
Each request is evaluated and sent to the model and provider that best fits task complexity, price, speed and reliability.
- ✓Automatic provider fallback
When a provider is degraded or down, traffic is re-routed to another host of the same model to preserve availability.
- ✓Bring your own keys (BYOK)
Use existing provider contracts and committed spend through OpenRouter's routing layer, free up to a monthly allowance.
- ✓Data policy and residency controls
Zero prompt logging by default, zero-data-retention provider routing, and org-wide policies excluding specific providers or regions.
- ✓Programmatic key management and budgets
Per-key credit limits with automatic resets and unified billing let finance cap spend per team or per application.
Capabilities
Use Cases
- •Multi-model product architecture
Serve a cheap fast model for routine turns and escalate hard requests to a frontier model without shipping two integrations.
- •Model evaluation and migration
Benchmark a newly released model against the incumbent on live traffic before committing engineering effort to a provider migration.
- •Outage resilience for AI features
Keep a customer-facing assistant answering during a model vendor incident by failing over to an equivalent model elsewhere.
- •Centralised LLM cost governance
Give every internal team a rate-limited API key and see consolidated token spend per team, product and model in one place.
- •Compliance-constrained inference
Restrict a regulated workload to zero-data-retention providers in approved regions using policy toggles instead of custom infrastructure.
Ideal For
Best For
- ✓Standardising a company on one LLM endpoint while keeping the freedom to change underlying models per feature
- ✓Evaluating and benchmarking new open-weight and frontier models the week they ship, before committing to a provider contract
- ✓Adding cross-provider failover so an outage at a single model vendor does not take a production feature offline
- ✓Consolidating LLM spend and per-key budget caps across many teams and applications under one invoice
- ✓Routing sensitive workloads only to providers that honour zero-data-retention and excluding specific regions or vendors
Not Ideal For
- ✗Very high-volume single-model workloads, where the 5%–5.5% take rate is pure overhead against a direct provider contract that also offers committed-use discounts
- ✗Latency-critical applications that cannot absorb the roughly 25–40ms the extra routing hop adds on top of provider inference time
- ✗Teams that require a contractual uptime SLA with service credits and a named support path; independent reviewers document three outages of 35–50 minutes between August 2025 and February 2026 and no published credit regime
- ✗Organisations whose compliance posture forbids any third party sitting between the application and the model provider, regardless of zero-logging defaults
Integrations
Deployment
Market & Ratings
10M+ users and 250k+ applications (vendor-reported); named enterprise customers include NVIDIA, Zoom, AMD, Replit, Framer, Webflow and Lovable
Market Analysis
Pros
- ✓Genuinely removes per-provider integration work: one API key, one schema, one invoice for 500+ models
- ✓No markup on inference tokens, and BYOK lets large buyers keep their existing provider discounts
- ✓Fast access to newly released open-weight models, which practitioners on Hacker News cite as faster than AWS Bedrock and similar catalogues
- ✓Easy spend control — top up a fixed amount, set per-key limits, and cap the blast radius of a runaway agent
Cons
- ✗The 5%–5.5% take rate is widely described by practitioners as the first thing to cut once volume is material; Hacker News commenters call it rent-seeking at scale
- ✗Thin moat — multiple commenters characterise it as 'just an API proxy' that AWS, Azure, GCP or the model vendors could replicate, and Menlo Ventures notes 10+ competing routers launched in recent weeks
- ✗Independent reviewers document three outages of 35–50 minutes between August 2025 and February 2026, and during the February incidents the platform returned misleading '401 User not found' errors that sent developers debugging their own credentials
- ✗Privacy is only as good as the downstream provider — despite SOC 2 and zero-data-retention options, reviewers question how verifiable the no-training guarantee is across 80+ third parties
- ✗Roughly 25–40ms of added latency per request, and credits expire after 365 days
- ✗Ownership uncertainty: the Stripe acquisition announced 19 August 2026 has not closed and roadmap implications are unknown
Pricing
Pay-as-you-go
From $0 (provider rates + 5.5% on credit purchases)
- ✓Provider inference pricing passed through with no markup
- ✓5.5% fee ($0.80 minimum) on card credit purchases, 5% on crypto
- ✓BYOK free to $25,000/month of usage, then 5%
- ✓Free-model tier: 50 requests/day, 1,000/day after $10 in credits
Enterprise
Contact for pricing
- ✓SOC 2 compliance and GDPR-compatible infrastructure
- ✓Zero-data-retention routing and data residency controls
- ✓SSO, organisations and global policy enforcement
- ✓BYOK free to $200,000/month of usage, then 5%
- ✓Dedicated engineering contact, priority support and uptime commitments
OpenRouter does not mark up inference — you pay the provider's published token rate. Revenue comes from a 5.5% fee ($0.80 minimum) on credit-card credit purchases (5% for crypto) and a 5% fee on bring-your-own-key usage above a monthly allowance of $25,000 pay-as-you-go or $200,000 enterprise. Credits expire after 365 days. Enterprise terms, SSO and uptime commitments are quote-only, and the take rate is the single most common cost objection raised by high-volume users.
Security & Compliance
Sources
This page was written from 7 sources, 4 on domains other than openrouter.ai.
- 1.openrouter.ai — openrouter.aivendor
- 2.openrouter.ai — enterprisevendor
- 3.openrouter.ai — faqvendor
- 4.stripe.com — stripe agrees to acquire openrouter
- 5.news.ycombinator.com — item
- 6.menlovc.com — stripe to acquire openrouter why everyone is obsessed with m
- 7.ofox.ai — is openrouter reliable honest review 2026
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
NVIDIA Nemotron 3.5 Lightning
Open 30B mixture-of-experts model tuned for the high-volume execution layer of long-running agents
River AI
Token-metered LoRA fine-tuning and reinforcement learning on open-weight models you keep
Scaled Cognition
APT, a large action model trained to take actions instead of predicting text
Engram
A learned memory layer that makes AI actually know your organization — at up to 100x fewer tokens.