# OpenRouter
**OpenRouter** is a hosted AI gateway and model marketplace that gives developers a single OpenAI-compatible API for accessing large language models from many providers, with routing, failover, and unified billing handled by the service. It was started in early 2023 as the first LLM marketplace and, by its own description, grew into the largest AI gateway.[1](https://openrouter.ai/about/) In August 2026 the payments company Stripe agreed to acquire it for more than $8 billion in cash and stock.[2](https://www.axios.com/2026/08/17/stripe-openrouter-paypal)
| Key fact | Detail |
|---|---|
| Founded | Early 2023, as the first LLM marketplace[1](https://openrouter.ai/about/) |
| API surface | One OpenAI-compatible endpoint, `https://openrouter.ai/api/v1`, one API key, 400+ models across 70+ providers[3](https://openrouter.ai/blog/insights/model-routing/) |
| Scale | 600T+ monthly tokens, 8M+ developers, 80+ providers, 500+ models (company figures)[1](https://openrouter.ai/about/) |
| Routing | Two independent layers: model routing (which model answers) and provider routing (which provider serves it)[3](https://openrouter.ai/blog/insights/model-routing/) |
| Fees | 5.5% credit fee on top of underlying provider pricing[4](https://continuumcode.ai/guides/openrouter-vs-litellm/) |
| Funding | $164 million raised, including a $113M Series B in May 2026 led by CapitalG at about a $1.3 billion valuation[2](https://www.axios.com/2026/08/17/stripe-openrouter-paypal) • [5](https://techcrunch.com/2026/05/26/openrouter-more-than-doubles-valuation-to-1-3b-in-a-year/) |
| Acquisition | Agreed to be acquired by Stripe for more than $8 billion, announced August 19, 2026[2](https://www.axios.com/2026/08/17/stripe-openrouter-paypal) • [6](https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter) |
## What the service does
A developer calling OpenRouter uses the same request format as the [OpenAI API](https://www.edgepedia.org/openai-api), and OpenRouter fans the request out behind the scenes to whichever model and provider it selects.[3](https://openrouter.ai/blog/insights/model-routing/) This removes two practical burdens: maintaining separate integrations and API keys for each vendor, and writing retry logic when a provider rate-limits or fails. The company frames its mission as increasing what it calls AI neurodiversity, a marketplace in which each model's distinct strengths are visible and developers are not locked into one vendor.[1](https://openrouter.ai/about/)
Routing is organized as two independent decisions: which model answers the request, and which provider serves that model. Keeping the layers separate lets a team fix a problem at one layer, for example by reordering providers, without changing which model handles the work.[3](https://openrouter.ai/blog/insights/model-routing/)
## Reliability and failover
**Provider-layer failover** is automatic and on by default. A single model such as [Claude Sonnet 4](https://www.edgepedia.org/claude-sonnet-4).6 is often served by several providers; if the chosen provider returns a 5xx error or rate-limits, OpenRouter tries the next provider for the same model. This behavior is governed by `allow_fallbacks`, which defaults to true.[7](https://openrouter.ai/blog/insights/reliability-failover/)
The load-balancing rule that OpenRouter publishes has three parts: providers are ranked so that those without any notable outage over the previous 30 seconds come first, the cheapest stable option is chosen with a weighting based on one over price squared, and remaining providers serve as backups. In the documented worked example, a provider charging $1 per million tokens is roughly nine times more likely to be tried before one charging $3 per million, because the weighting goes as 1/3² = 1/9; a provider that recently suffered outages drops to the back of the line without being excluded.[7](https://openrouter.ai/blog/insights/reliability-failover/)
**Model-layer fallbacks** are opt-in through a `models` array. If every provider for the primary model fails, or the request hits a context-length error or a moderation refusal that no alternative provider can fix, OpenRouter moves to the next model in the caller's list. This layer is opt-in because it changes which model answers, a decision the company leaves to the user.[7](https://openrouter.ai/blog/insights/reliability-failover/)
Callers can also shape the candidate set directly: `order` tries providers in an explicit sequence, `only` is an allow-list, `ignore` is a block-list, and `allow_fallbacks: false` hard-stops at the chosen providers.[7](https://openrouter.ai/blog/insights/reliability-failover/)
## Provider network
Inference providers join by integrating their API with the OpenRouter network. Requirements include an OpenAI-compatible `/chat/completions` endpoint with streaming and usage token counts, a published `/models` endpoint listing pricing, context length, supported features, and datacenter locations, monthly invoicing, and a published privacy policy disclosing whether prompts are logged or used for training.[8](https://openrouter.ai/providers/apply)
Routing to providers is based on price, latency, throughput, and reliability, with high-performing providers receiving proportionally more traffic. Providers maintaining 95% or better uptime retain standard routing priority; between 80% and 94% traffic is reduced, and below 80% an endpoint is used only as fallback. For tool-calling requests, the company's Auto Exacto system further optimizes routing based on tool-call success rates. Time to first token, throughput, and uptime are tracked publicly on each model page.[8](https://openrouter.ai/providers/apply)
## Pricing and privacy
OpenRouter uses a credit system denominated in US dollars and charges a 5.5% credit fee on top of the underlying provider prices.[4](https://continuumcode.ai/guides/openrouter-vs-litellm/) • [9](https://openrouter.ai/support/) The company states that it passes through the pricing of the underlying model providers without markup.[9](https://openrouter.ai/support/)
On privacy, OpenRouter logs basic request metadata such as timestamps, model, and token counts, and states that prompts and completions are not logged by default. Users can opt in to prompt logging for a small discount, and a Zero Data Retention policy option is available. Requests are proxied to model providers, and routing respects the user's privacy settings; if no provider matches those settings, the request errors rather than being routed anyway.[9](https://openrouter.ai/support/) Provider behavior still matters: each provider's own terms govern whether it logs prompts or trains on them, and model pages on the marketplace carry logging notices for individual models.[10](https://openrouter.ai/openrouter)
## The Auto Router and cloaked models
OpenRouter operates its own models alongside the marketplace. The **Auto Router** automatically selects a model for a prompt based on what the OpenRouter community collectively spends on similar tasks, looking at spend over a trailing 7-day window so it tracks new releases; the response is priced at the routed model's rate.[10](https://openrouter.ai/openrouter)
The platform has also hosted a series of **cloaked models**, anonymous frontier models released under codenames such as Quasar Alpha, Optimus Alpha, Sonoma Sky Alpha, and Horizon Alpha so the community could test them and give feedback before an official launch. Several were later revealed: Polaris Alpha as an early snapshot of GPT-5.1, the Sherlock models as early snapshots of [Grok 4](https://www.edgepedia.org/grok-4).1 Fast, and Bert-Nebulon Alpha as an early testing version of [Mistral Large 3](https://www.edgepedia.org/mistral-large-3).[10](https://openrouter.ai/openrouter)
## Growth and funding
OpenRouter raised a $40 million Series A in June 2025 led by [Andreessen Horowitz](https://www.edgepedia.org/andreessen-horowitz) and [Menlo Ventures](https://www.edgepedia.org/menlo-ventures) with Sequoia participation, at an estimated $547 million post-money valuation.[5](https://techcrunch.com/2026/05/26/openrouter-more-than-doubles-valuation-to-1-3b-in-a-year/) In May 2026 it raised a $113 million Series B led by CapitalG, Alphabet's independent growth fund, with participation from NVentures (NVIDIA's venture arm), ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures, Databricks Ventures, AMP PBC, and [Pace Capital](https://www.edgepedia.org/pace-capital), at a reported valuation of about $1.3 billion.[5](https://techcrunch.com/2026/05/26/openrouter-more-than-doubles-valuation-to-1-3b-in-a-year/) • [11](https://openrouter.ai/blog/announcements/series-b/) Total funding reached $164 million.[2](https://www.axios.com/2026/08/17/stripe-openrouter-paypal)
Usage grew quickly alongside the funding. The company reported that weekly volume rose from 5 trillion to 25 trillion tokens in the six months before the Series B, with 8 million or more developers building across 400+ models, and stated it had seen at least 10x growth in inference volume every year since founding.[11](https://openrouter.ai/blog/announcements/series-b/) Sacra, a private-market research firm, estimated annualized revenue of about $50 million in early 2026, up from roughly $10 million in October 2025.[12](https://sacra-pdfs.s3.us-east-2.amazonaws.com/openrouter.pdf)
## Acquisition by Stripe
On August 19, 2026, Stripe announced it had agreed to acquire OpenRouter, describing it as a leading AI model gateway and routing platform that helps businesses route and optimize token usage across 400+ models from more than 80 providers.[6](https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter) Axios reported the price as more than $8 billion in cash and stock, and noted that Stripe already processes payments for frontier AI labs, so the acquisition gives it a more direct role in how developers evaluate, select, and move between models.[2](https://www.axios.com/2026/08/17/stripe-openrouter-paypal) OpenRouter's own announcement said it was processing more than 10 trillion tokens per day at the time.[13](https://openrouter.ai/blog/announcements/openrouter-is-joining-stripe/)
## Research use and ecosystem role
Because OpenRouter sits between many developers and many providers, its traffic data has become a window into real-world LLM use. A December 2025 study, *State of AI: An Empirical 100 Trillion Token Study with OpenRouter*, analyzed over 100 trillion tokens of interactions on the platform and reported substantial adoption of open-weight models, outsized popularity of creative roleplay alongside coding assistance, and the rise of agentic inference; the authors noted that the study reflects a single platform over a finite window and so offers only a partial view of the ecosystem.[14](https://arxiv.org/abs/2601.10088)
Marketplace data also informs industry analysis. A Mozilla report counted eight of the top ten models by August token volume on OpenRouter as open weights, seven of them Chinese-built, while the [Linux Foundation](https://www.edgepedia.org/linux-foundation) reported that closed providers still took 96% of model-layer revenue on the platform from May to September 2025.[15](https://www.tomshardware.com/tech-industry/artificial-intelligence/chinas-open-weight-ai-models-are-now-just-4-months-behind-frontier-us-offerings-mozilla-report-claims-models-still-lag-in-some-benchmarks-but-are-drastically-cheaper-to-use)
**Comparison with self-hosted gateways.** OpenRouter's main alternative is LiteLLM, an open-source SDK and self-hosted proxy. The tradeoff is managed convenience against control: OpenRouter offers zero-ops access, unified billing, and broad model reach, while requests flow through a third party and the service does not hold independent [SOC 2](https://www.edgepedia.org/soc-2), ISO 27001, or HIPAA certifications, so a deployment with strict data-residency rules is responsible for meeting those standards itself.[4](https://continuumcode.ai/guides/openrouter-vs-litellm/) • [16](https://openrouter.ai/blog/insights/openrouter-vs-litellm/) A common pattern is a self-hosted LiteLLM proxy with OpenRouter configured as one upstream provider, combining LiteLLM's governance and budget controls with OpenRouter's model breadth.[16](https://openrouter.ai/blog/insights/openrouter-vs-litellm/)
## References
1. [About OpenRouter](https://openrouter.ai/about/)
2. [Stripe strikes mega-deal for OpenRouter, Axios](https://www.axios.com/2026/08/17/stripe-openrouter-paypal)
3. [How OpenRouter Model Routing Works, OpenRouter Blog](https://openrouter.ai/blog/insights/model-routing/)
4. [OpenRouter vs LiteLLM: fees, ops, privacy, and coverage, Continuum Code](https://continuumcode.ai/guides/openrouter-vs-litellm/)
5. [OpenRouter more than doubles valuation to $1.3B in a year, TechCrunch](https://techcrunch.com/2026/05/26/openrouter-more-than-doubles-valuation-to-1-3b-in-a-year/)
6. [Stripe agrees to acquire OpenRouter to help businesses optimize token routing and usage, Stripe Newsroom](https://stripe.com/newsroom/news/stripe-agrees-to-acquire-openrouter)
7. [OpenRouter Failover and Fallbacks Explained, OpenRouter Blog](https://openrouter.ai/blog/insights/reliability-failover/)
8. [Become a Provider, OpenRouter](https://openrouter.ai/providers/apply)
9. [Support, OpenRouter](https://openrouter.ai/support/)
10. [OpenRouter API and Models](https://openrouter.ai/openrouter)
11. [OpenRouter Raises $113M Series B, OpenRouter Blog](https://openrouter.ai/blog/announcements/series-b/)
12. [OpenRouter company profile, Sacra](https://sacra-pdfs.s3.us-east-2.amazonaws.com/openrouter.pdf)
13. [OpenRouter is Joining Stripe, OpenRouter Blog](https://openrouter.ai/blog/announcements/openrouter-is-joining-stripe/)
14. [State of AI: An Empirical 100 Trillion Token Study with OpenRouter, arXiv](https://arxiv.org/abs/2601.10088)
15. [China's open-weight AI models are now just 4 months behind frontier US offerings, Mozilla report claims, Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/chinas-open-weight-ai-models-are-now-just-4-months-behind-frontier-us-offerings-mozilla-report-claims-models-still-lag-in-some-benchmarks-but-are-drastically-cheaper-to-use)
16. [OpenRouter vs LiteLLM: Managed vs Self-Hosted Gateway, OpenRouter Blog](https://openrouter.ai/blog/insights/openrouter-vs-litellm/)
---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › AI companies, people, and products › AI chips, compute, and infrastructure companies*
*Initially written Sep 17, 2026 · Reviewed: Sep 30, 2026; Oct 15, 2026 · Edited: Sep 30, 2026; Oct 11, 2026; Oct 15, 2026 · Last review: Oct 15, 2026*
*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*
License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license