What Is OpenRouter? Pricing, Fees, Free Models, and When Another Gateway Is Better

OpenRouter is a gateway service (router) that lets you call hundreds of AI models through one API. Because it provides an OpenAI-compatible API, you only need to point the OpenAI SDK's base_url to https://openrouter.ai/api/v1 and specify provider-qualified model IDs such as anthropic/… or openai/… to use models from different companies with one key. Pricing uses the provider's model price with no markup, while purchasing credits incurs a 5.5% fee (minimum $0.80) (cryptocurrency payments incur 5%). :free-suffixed free models are also available, with a limit of 50 requests per day (or 1,000 per day if you have purchased at least $10 cumulatively).

All statements about OpenRouter on this page are based on OpenRouter's own public documentation (including its FAQ, pricing page, and rate-limit page, as of September 2026), with sources linked at the relevant points. The later section, “Cases where another type of gateway is a better fit,” is a comparison from Kunavo's perspective as the operator of this site. Since it concerns our own product, read it with that context in mind.

How OpenRouter works

Users have one OpenRouter account and choose the model by name for each request. As described in OpenRouter's quickstart, existing OpenAI SDK code works by changing only base_url and the key.

When multiple providers offer the same model, OpenRouter distributes load among higher-ranked providers by default to maintain availability (Provider Routing). If you list alternative models in the models parameter, it automatically tries the next one when the original model is down or rate-limited, and charges according to the price of the model actually used (Model Fallbacks).

The API is centered on chat completions, but that is not all it offers. There is a dedicated image generation API and an asynchronous video generation API, and instructions for using it from Claude Code through Anthropic-format endpoints are also available. The description “OpenRouter is text-only” is no longer accurate as of September 2026.

OpenRouter pricing and fees

The key point is: “model prices are the same as the provider’s, and fees apply when purchasing credits.” The following summarizes the information in the OpenRouter FAQ and pricing page (as of September 2026).

Because the fee has a minimum amount, smaller top-ups result in a higher percentage. For a $10 top-up, the fee is the minimum amount of $0.80 (8%), while for $100 it is $5.50. Prices may change, so check the latest information at the links above before signing up.

How to use free models and their limits

Append :free to the end of a model ID to call its free version. According to the rate limits page, the limits are 20 requests per minute and 50 per day, increasing to 1,000 per day if you have purchased at least $10 in cumulative credits. The documentation also states that free versions may have different rate limits and availability from paid versions.

They are sufficient for trying things out, learning, and small personal tools, but the daily limit makes them unsuitable for processes that cannot stop. Coding agents such as Cline call a model dozens of times for a single task, so the free allowance runs out quickly.

How to use OpenRouter

1. Create an account at openrouter.ai.
2. Purchase credits if you plan to use paid models (not necessary if you use only free models).
3. Create an API key.
4. Point base_url in the OpenAI SDK to https://openrouter.ai/api/v1, then call it with a provider-qualified model ID.

import os
from openai import OpenAI

# OpenRouter — モデル ID は「提供元/モデル名」の形
openrouter = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key=os.environ["OPENROUTER_API_KEY"],
)
r = openrouter.chat.completions.create(
    model="anthropic/claude-sonnet-4.6",
    messages=[{"role": "user", "content": "OpenRouter とは何か、1 文で"}],
)

# 無料モデルは ID の末尾に :free（例は OpenRouter のドキュメントのもの。
# 提供状況と 1 日の回数制限は OpenRouter 側で変わる）
free = openrouter.chat.completions.create(
    model="meta-llama/llama-3.2-3b-instruct:free",
    messages=[{"role": "user", "content": "こんにちは"}],
)

# Kunavo — 変わるのは base_url、キー、モデル ID の書き方の 3 つだけ
kunavo = OpenAI(
    base_url="https://api.kunavo.com/v1",
    api_key=os.environ["KUNAVO_API_KEY"],   # sk-kn-...
)
r = kunavo.chat.completions.create(
    model="claude-sonnet-5",
    messages=[{"role": "user", "content": "OpenRouter とは何か、1 文で"}],
)

As the final three lines show, when moving between OpenAI-compatible gateways, only base_url, the key, and the model ID format change. OpenRouter adds a provider prefix, as in anthropic/claude-sonnet-4.6, while Kunavo does not, as in claude-sonnet-4-6. From creating a key through making the first call, see the Kunavo quickstart.

Where OpenRouter is strong

• Model breadth. You can choose from hundreds of models, including open-source derivatives, through one API. If you need long-tail models, there are almost no alternatives to OpenRouter.
• Free models. There are at least 25 free models that you can try at no cost within the daily limit.
• The same model from multiple providers. Load balancing among providers and automatic failover to alternative models work without configuration.
• Bring your own key. You can use keys from contracts made directly with providers through OpenRouter, with no fees up to $25,000 per month.
• Cryptocurrency payments. USDC payments are supported.

If your main goal is broad coverage of text models and free models, staying with OpenRouter is reasonable. There is no compelling reason to switch.

When a different kind of gateway is a better fit (Kunavo’s perspective from here)

OpenRouter uses the model provider's list price as-is plus a fee at purchase. Kunavo uses the opposite approach, selling at per-model unit prices discounted from the provider's list price. Top-ups start at $10, giving a $10.00 balance, and top-ups of $100 or more receive a bonus ($100 → $110.00, $1,000 → $1,200.00, $5,000 → $6,250.00).

Discounts vary by model. Comparing monthly costs, using 10 million input tokens and 1 million output tokens with Claude Opus 5 costs $75.00 at the provider's list price (if OpenRouter charges the same rates as the provider, as stated in its FAQ, a purchase fee is added, bringing the total to approximately $79.13), compared with $52.50 on Kunavo. Rates for all models are available on the pricing page, and a model-by-model comparison for Claude is available under Claude pricing.

There are three differences beyond unit prices.

• Images, video, and music with the same key. OpenRouter also has image and video APIs, but Kunavo’s image API uses the same format as OpenAI’s Images API (/v1/images/generations and /v1/images/edits), so images.generate in the OpenAI SDK works as-is. Music generation (Suno) can also be called with the same key. As of September 2026, we could not find a music-generation page in OpenRouter’s documentation.
• Easy to pay from Japan. Stripe Checkout accepts cards including JCB (Visa, Mastercard, American Express, JCB, UnionPay), Apple Pay, Google Pay, and Link. Convenience-store payments and PayPay are not supported.
• Balances do not expire. There is no charge for months in which you do not use the balance, and failed requests are not charged.

And to be clear about Kunavo's disadvantages: it runs on shared capacity, with no dedicated quota and no contractual SLA. If you need guarantees, contract directly with the relevant providers. The model selection is limited to major current models, so there is no long tail, no free models, and cryptocurrency (stablecoin) payments are limited to US-dollar checkouts for the United States and cannot be used from Japan.

A side-by-side comparison by category is available in Kunavo vs. OpenRouter (English), while OpenRouter alternatives (English) lists options beyond OpenRouter.

Frequently asked questions