The free tier is most easily misunderstood as having "25+ free models", which sounds like a lot, but they all share a quota of 50 requests per day, which can be used up quickly with just a little testing. If you actually want to use it for work, you still need to pay for more.
View Change History
2026-08-03~2026-10-06Routine checks ran 28 times; no price or plan changes
2026-09-25~2026-10-05Automated checks could not be matched line by line with the official site 9 times (page redesign, region-based pricing or usage-based billing); not judged to be a price change; the verified date was not updated, pending manual review
Continuously re-checked, every fact dated
Has a free plan
What Is This
One account, 400+ models. Automatic price routing and failover across 70+ providers.
Free Version Limitations
More than 25 free models, 4 free providers, daily limit of 50 requests, community support
Available Models And Quantity
Model
Free quota
Calculate Time Window
Description
25+ free models (provided by each vendor)
50 requests Daily
Resets daily
The list of free models is subject to change based on the supplier's policy, and the actual number of available models is determined by the official website at the time.
Free Usage Limit
Features
Free quota
Description
Number of free models
25+ items
From 4 free providers
Daily request limit
50 times
This is the main limitation of the free tier
Paid model access
No
Requires pre-paid points
Auto Price Comparison Route
Available
When the same model is available from multiple suppliers, it will automatically choose the cheaper or available one.
Best for: Those who want to use GPT, Claude, Gemini, and open-source models at the same time without subscribing to multiple monthly fees.
API pay-as-you-go (usage-based)
No monthly fee, no minimum consumption. A 5.5% platform handling fee is charged when purchasing points, and the model itself is billed at the original price of each supplier. It supports 500+ models, 80+ suppliers (updated on 2026-08-22: the scale has grown compared to previous records). It supports credit cards and cryptocurrencies (as of 2026-07-31, according to the official website openrouter.ai/pricing). BYOK (bring your own key) quota: within US$25,000 list price inference per month is free of handling fees, and 5% is charged for excess amounts.
Which Model
400+ models, 70+ providers, switchable at will
Usage Quota
Based on the prepaid points, deductions are made according to usage, with no monthly fee and no minimum consumption.
Max Reading Time
Depends on the selected model
This Plan Includes
400+ models in a single API
Automatic price comparison routing (choosing the cheapest or fastest provider)
Supplier outage automatic rerouting
Usage analysis
Credit card and cryptocurrency payment
Not Included
Platform fee discount
Dedicated rate limit
SLA and Dedicated Support
What Sets Us Apart
The biggest difference from a subscription model is **no monthly fee**. If you only use high-end models occasionally in a month, pay-per-use is usually more cost-effective than paying a US$20 monthly fee; however, note that a 5.5% platform fee will be added when topping up.
Best for: Teams or companies with large usage requirements that need customized pricing and service guarantees
API pay-as-you-go (usage-based)
Negotiable platform fee discounts, dedicated rate limits, SLA, and shared Slack channel; BYOK (bring your own key) quota: within US$200,000 list price per month for inference, no handling fee, 5% charge for excess (this is for BYOK Limits, not general usage)
Which Model
Same as Pay-as-you-go, and also customizable configuration available
Usage Quota
Monthly list price up to US$200,000 is exempt from handling fees, with a 5% charge on the excess amount.
Max Reading Time
Depends on the selected model
This Plan Includes
Platform fee discounts are negotiable
Dedicated rate limit
SLA Service Guarantee
Shared Slack support channel
What Sets Us Apart
It's only cost-effective for large-scale usage. For individuals and small teams, Pay-as-you-go is sufficient.
Model aggregator: does not develop models itself, but aggregates 400+ models under one API/account, allowing for automatic selection of the cheapest or fastest provider, and automatic switching to other providers if a single provider goes down. The reason for the partial zh_support label is: the platform interface is primarily in English, but can directly call Chinese-ability-strong models; the interface localization degree requires manual verification. Whether the daily limit of 50 times for the free quota includes all free models also requires verification.
Related Guide: Five Brakes to Set Before an AI Agent Touches Money: A Checklist Derived From the US$12,431 Invoice Incident
A third-party experiment showed agents rerouting through a payments tool once email limits blocked them, and rationalising the crossing in their own reasoning traces. Seven agents burned about US$3,200 in three days, nine tenths of it on inference. Here are those failure modes turned into five brakes you can set today: hard spend caps, sandboxed payments, allowlisted recipients, stop conditions, and auditable traces. Last verified: 2026-09-08
Looking for OpenRouter deals, discounts or promo codes? We check the official site automatically every day, review changes by hand, and date-stamp every entry.
