What Is OpenRouter? A Guide for Australian Businesses

Blue ethernet cables plugged into a row of ports on a network switch, with green and amber status lights glowing.

OpenRouter is a service that gives your business one account, one bill and one API key for hundreds of AI models from different companies, so you can switch between models or use several at once without signing up to each provider. It has been getting a lot of attention from Australian developers and business owners this month, and new data from Databricks helps explain why: more organisations are moving away from relying on a single AI provider.

Here’s what OpenRouter is, what changed recently, what it costs, and the privacy questions to answer before you send it real work.

What is OpenRouter?

OpenRouter sits between your app and the AI companies. Instead of holding separate accounts with OpenAI, Anthropic, Google, DeepSeek, Mistral and others, you send every request to OpenRouter and name the model you want. According to the OpenRouter FAQ, it provides “a unified API to access all the major LLM models on the market”, with billing and usage tracking in one place.

A few things make it different from going direct:

  • One format for every model. OpenRouter follows the OpenAI API format, so tools and code built for OpenAI can usually point at OpenRouter with a changed address and key.
  • Automatic fallbacks. If a provider returns an error, OpenRouter says it automatically tries the next provider. You can also list backup models in priority order.
  • A chat screen as well as an API. You can compare models in its web chat before building anything.

OpenRouter is a United States company. In August it announced it is joining Stripe, the payments company, and said its product, name and roadmap would stay the same.

Why OpenRouter is in the news this month

  • Databricks AI Index (published 16 November, US time). Databricks launched a public AI Index built from anonymised data on companies that use its Unity Gateway, its tool for accessing and governing AI models. It reports that about 64% of those companies relied on just one model provider in July, falling to 28% by 2 November. That “use more than one model” shift is what OpenRouter is built for.
  • A new Auto Router (6 November). OpenRouter rebuilt its Auto Router, which picks a model for each request for you, and added Auto Code, a version tuned for coding tools. The Auto Router documentation says it only chooses from models your spending, privacy and model rules allow.
  • Hugging Face (16 November, US time). OpenRouter announced that chat models on Hugging Face Inference Providers now run through OpenRouter’s routing, which picks a provider for each request based on live price, speed and reliability.
  • A new data tier (26 October). OpenRouter introduced Safety Retention, where OpenRouter or a provider may store requests to detect and prevent abuse, but not to train models. More on why that matters below.

Read the open-weight figure carefully

One Databricks figure is being shared on social media as open models now making up “over 30%” of enterprise AI. The source is narrower. The chart note on the Databricks AI Index page says the proprietary versus open-weight split counts tokens sent to Databricks-hosted models on its Unity Gateway, by Databricks customers. On that measure, Databricks reports the open-weight share (models such as GLM, Kimi, DeepSeek, Qwen, Llama, GPT-OSS and Gemma) rose from 7.3% in July to 30.9% in its reading for the week of 2 November.

It’s a useful signal, but it describes one platform’s customers, not all businesses, and the same applies to the single-provider figures. Databricks notes results “may vary across organizations”.

Why businesses are using more than one AI model

The practical reasons:

  • Different jobs suit different models. A cheap, fast model may be fine for sorting emails or tagging records, while a more capable model is worth paying for when drafting a tender response or reviewing code.
  • Price per token isn’t the full cost. Databricks found that cost per completed task depends on how much work a model does to reach an answer, so a model with a low token price can still end up expensive. Test on your own tasks, not on price lists.
  • Models change fast. Being able to switch without rebuilding your app reduces the risk of being stuck with one supplier’s pricing, limits or outages.
  • Resilience. If one provider is down or rate-limiting you, a fallback model can keep a customer-facing tool working.

Our free choosing AI tools for business course walks through matching tasks to model tiers, reading AI pricing and running a fair trial.

What OpenRouter costs

OpenRouter works on prepaid credits. As at today, its FAQ says:

  • Model prices are passed through. OpenRouter says you pay the same rate as you would directly with the provider, with no markup on usage.
  • There is a fee when you buy credits. OpenRouter charges 5.5% (minimum US$0.80) on credit purchases. If you bring your own provider keys, there is a monthly free allowance on pay-as-you-go, then a fee on usage above it.
  • Everything is in US dollars. Check whether your card adds foreign currency fees.
  • Credits have conditions. Unused credits can be refunded only within 24 hours of purchase, and OpenRouter reserves the right to expire unused credits after one year.
  • Free models exist, with low limits. OpenRouter says free models have low daily rate limits and “are usually not suitable for production use”.

For most businesses the benefit isn’t a cheaper token price. It’s managing one account instead of several, and being able to move routine work to a cheaper model once you’ve tested it.

OpenRouter, privacy and Australian data

This is where Australian businesses need to slow down. OpenRouter’s data collection page says it does not store your prompts or responses by default, unless you opt in to logging, content is flagged for child-safety screening, or the request goes to a Safety Retention endpoint (see below). Your request still goes on to whichever provider serves the model, and providers have their own policies.

  • Three data tiers. Endpoints are labelled Zero Data Retention (no routine storage), Safety Retention (may be stored for abuse prevention, not training) or Retention and Training (may be stored and used for training under the provider’s policy). You can restrict your account to Zero Data Retention only.
  • Training is off by default. OpenRouter says providers that log, or whose policy it can’t confirm, aren’t used unless you switch on a training setting. There is also an opt-in that lets OpenRouter use your prompts for a small discount. Leave both off for business data.
  • No Australian region. OpenRouter’s in-region routing keeps prompts in the EU or the US, and only on its Business and Enterprise plans. It doesn’t list an Australian option. Its docs also note that account, billing and operational records may be stored or processed in the US.

Under the Privacy Act, sending personal information to an overseas service raises cross-border disclosure obligations under APP 8. The OAIC’s guidance on commercially available AI products says privacy obligations apply to personal information put into an AI system and to outputs that contain it, and recommends as best practice that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. If your data must stay onshore, compare our private LLM options guide and our free local LLM for business course.

When OpenRouter makes sense, and when it doesn’t

It can be a good fit if you:

  • want to compare several models on your own tasks before committing to one;
  • are building an internal tool, prototype or AI agent and want backup models if one provider fails;
  • use developer tools that already support OpenRouter, such as the open-source agent we covered in What Is Hermes Agent?;
  • want to try open-weight models like Mistral Large 4 without renting your own servers.

Think twice if you:

  • handle health, legal, financial or other sensitive personal information;
  • have contracts or industry rules that require Australian data residency;
  • already have an enterprise agreement with one provider that covers your needs;

How to trial OpenRouter safely

  1. Pick one low-risk task, such as summarising public documents or drafting product descriptions, with no customer or staff personal information.
  2. Set your privacy settings first. Decide whether to enforce Zero Data Retention and whether to allow Safety Retention endpoints, and leave logging and training off.
  3. Set a budget. OpenRouter’s guardrails let you cap spending daily, weekly or monthly, and limit which models and providers each key can use.
  4. Protect the API key. One key unlocks every model, so keep it out of code and shared documents, and give each app its own key. Our post on LLMjacking and stolen AI API keys explains why.
  5. Compare two or three models on the same set of real examples, and note quality, time and cost for each.
  6. Decide and document. Record which model handles which task, why, and when you’ll review it.

Caveats before you sign up

  • Another party in the chain. OpenRouter adds a supplier between you and the model provider, so you need to trust both and read both sets of terms.
  • Automatic choices need watching. Routers and fallbacks can send a request to a different model or provider than you expect. Check your activity log, and use allowlists if consistency matters.
  • Benchmarks aren’t your business. Databricks’ coding results come from its own codebase. Your results will depend on your tasks.
  • Things change quickly. Fees, data tiers and model lists can change at short notice.

If you’re still working out which AI tools to use at all, start with our free AI tools for small business course.

This is general information based on OpenRouter’s documentation and announcements, the Databricks AI Index and OAIC guidance, as at 17 November 2026. OpenRouter’s pricing, data policies and features change often and may vary by plan. It is not legal or privacy advice; get advice on your own obligations before sending personal information to any AI service.

Thinking about OpenRouter or more than one AI model?

Start with our free choosing AI tools for business course to match tasks to models and keep spending in check. If you’d like a second opinion, an AI business audit shows where AI fits in your operations. Aideveloper designs and builds LLM integrations for Australian businesses, including model choice, fallbacks, privacy settings and spend controls. Talk to us about the right setup for your data.

Featured image: Photo by Brett Sayles on Pexels.