FlyRouter
Direct access to global models · Up to 90% off

Agent gateway
One API for every top AI model

Read the docsOpenAI-compatible

https://api.flyrouter.ai/v1

Direct access to global models

ClaudeClaude
OpenAIGPT
GeminiGemini
DeepSeekDeepSeek

Smart routing

AGENT CLIENTS

Claude CodeClaude Code
CodexCodex
CC Switch
OpenClawOpenClaw

10+ agents and clients supported

  • Smart routing
  • Prompt caching
  • Protocol passthrough
  • One bill
  • Usage tracking
  • Never downgraded

Live pricing · up to 90% off official

Prices update live from the gateway. Strikethrough prices are official rates. Each request is charged to your balance based on actual usage.
Text models are priced in USD / 1M Tokens. Image and video models are priced per spec in USD / image, USD / sec or USD / video; per-request models in USD / request.

Billing rules  →

Quick start

Compatible with major API protocols. Change one line, the Base URL, and you are ready for production;clients like Codex and Claude Desktop set up in minutes.

Also works with Codex, Claude Desktop, OpenClaw, CC Switch, and any OpenAI-compatible client.

View API docsGo to production: authentication, endpoints, and error codesConnect agent clientsSet up Codex, Claude Desktop, and other dev tools
1from openai import OpenAI2 3client = OpenAI(4    base_url="https://api.flyrouter.ai/v1",  # The only line you need to change5    api_key="<FLYROUTER_API_KEY>",6)7 8response = client.chat.completions.create(9    model="gpt-6.1-sol",10    messages=[{"role": "user", "content": "Hello"}],11)12 13print(response.choices[0].message.content)

Point your Base URL at FlyRouter, add your API key, and leave the rest of your code as is.

A routing layer you can trust in production

Leading frontier models in one place. Every model price can be checked on this page.

Models available

—

The number of models you can call. Input and output prices for every model are published on this page.

Avg. time to first token

—

Average wait from sending a request to the first token over the last 24 hours.

Request success rate

—

Share of requests that completed successfully over the last 24 hours, across all models.

Why FlyRouter?

Run agent workloads reliably at low cost, with model behavior and spend that stay stable and auditable.

  • No downgrades

    Requests go straight to verified upstreams, never silently swapped or downgraded.

  • Data privacy

    Prompts and outputs are used only for routing, metering, and billing. Never trained on, never sold.

  • Traceable spend

    Every request's model, token usage, and charge is listed line by line.

How do we get these prices?

One account, one bill. Every upstream provider is vetted by us.

  • Multi-provider sourcing

    We work with multiple upstream providers, vet each one for quality, and buy from the best, so you don't have to shop around.

  • Bulk purchasing

    We buy credits in bulk from vetted providers and pass the discount straight to you.

  • No plans required

    Pay as you go. Your balance never expires, with no monthly fee and no minimum spend.

How a request flows through the gateway

The gateway never rewrites requests: whatever model and protocol you call is forwarded as-is to that model's configured channel. Authentication, groups, quota, and metering all happen before forwarding, and if your balance runs out you get an error, not a cheaper model in disguise.

Protocols
Native OpenAI, Anthropic, and Gemini protocols passed through as-is
Price
Up to 90% off official prices, charged per request by actual usage
Errors
Upstream errors returned honestly, never a silent model swap or downgrade
Billing
Model, token usage, and charge listed per request, reconcilable line by line
  1. 01

    Client

    Claude Code, Codex, or any OpenAI-compatible client. Change only the Base URL, nothing else.

    https://api.flyrouter.ai/v1
  2. 02

    Auth and quota

    Checks your API key, group, and balance. If credits run out, you get an error, never a rerouted model.

    Authorization: Bearer sk-…
  3. 03

    Forward to channel

    Finds the upstream channel for the requested model and forwards the body, model name, and protocol as-is.

    POST /v1/chat/completions
  4. 04

    Metering and response

    Streams responses through as they arrive, then bills by the upstream usage. Every charge is auditable.

    data: [DONE]

Each model currently runs on one primary route. Upstream errors are returned as-is, with no silent model swaps or downgrades

Make every token go further

Pay only for what you use. Credits never expire and are deducted per request in real time,so your budget lasts longer.

Credits never expire·Top up anytime

FAQ

Answers to common questions

How do I get started?

Sign up, top up in the console, and create an API key. Then point the Base URL in your code or client to https://api.flyrouter.ai/v1 and add your key. Nothing else in your code needs to change.

What is an LLM router?

A unified API that sits between your tools and model providers. You call one endpoint with one key; the gateway forwards each request to the channel for the model you choose, meters usage, and gives you a single consolidated bill.

Is FlyRouter OpenAI-compatible?

Yes, and it supports both formats. The OpenAI-compatible endpoint works with the official OpenAI SDK and Codex by changing only the base URL. Claude Code connects through the Anthropic-compatible endpoint with two environment variables.

Can I use it with Claude Code, Codex, and other clients?

Yes. Claude Code connects through the Anthropic-compatible endpoint and Codex through the OpenAI-compatible endpoint. Clients like CC Switch and OpenClaw only need the Base URL and your key. See the API docs for step-by-step setup.

Which models are supported?

Leading frontier models including Claude, GPT, and Grok, with more added over time. The full list, live prices, and official prices are all on the pricing page.

How can prices be up to 90% off?

The discount comes from wholesale pricing: we buy upstream capacity in bulk and resell it below official rates, passing the difference on to you. Discounts vary by model, and the pricing page shows current and official prices side by side so you can check.

How does billing and top-up work?

You pay for actual usage, priced in USD. Input, output, and cache prices for each model are listed on the pricing page. You can top up in the console with Alipay or WeChat Pay, and your balance is credited instantly.

Do credits expire?

No. Top up anytime and use credits whenever you like. No subscription required.

Am I charged for failed requests?

Requests that fail because of an upstream outage are not charged. Every call's model, token usage, and charge is logged in the console so you can check them line by line.

Can I get a refund?

Unused balance can be refunded to the original payment method within 7 days of top-up; contact us to request it. Payment processing fees are deducted, and used credit is non-refundable.

Will my data be used to train models?

No. Prompts and completions are used only for routing, metering, billing, risk control, and support. They are never used for training or sold as usage data.

How can I check whether the service is up?

The status page shows each model's success rate and latency over the last 24 hours. If something looks off, check there first to see whether it's a wider outage.

Need help or want to share feedback?

Email support@flyrouter.ai, or reach us through the other channels on the Contact page. We'll get back to you as soon as we can.

FlyRouter

One key, every top AI model