virouter Create account
Priority · Claude Opus 4.8 · OpenAI + Anthropic

One key.
Two frontier
labs.

Virouter is a routing switchboard for developers who build with OpenAI and Anthropic. Use GPT and Claude behind one API key, one shared token wallet, and one OpenAI-compatible endpoint. Bills in tokens, not dollars — so you ship without provider-account chaos.

  • One API key
  • One token wallet
  • One OpenAI-compatible URL
  • OpenAI + Anthropic only
Create free account · 100k tokens → Read the docs

Free to start · No card · Works with the OpenAI SDK you already have.

Latency p50
182 ms
Routes / sec
9.4k
Uptime · 90d
99.98%
FIG. 00 · Drop-in

Keep your SDK.
Swap one URL.

If you already use the OpenAI SDK, point baseURL at Virouter. Same client, same parameters — now also routed to Claude when you ask for it.

  • Same SDK. OpenAI client and Anthropic client both work; no new library to learn.
  • One wallet. GPT and Claude calls deduct from the same token balance, visible in the dashboard.
  • Auto failover. When upstream returns 5xx, the relay retries the next healthy lane in <90ms.
node · openai sdk drop-in 
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.virouter.com/v1",
  apiKey:  process.env.VIROUTER_API_KEY,
});

// GPT
await client.chat.completions.create({
  model: "gpt-5.5",
  messages: [{ role: "user", content: "Hello" }],
});

// Claude — same client, same wallet
await client.chat.completions.create({
  model: "claude-opus-4-8",
  messages: [{ role: "user", content: "Hello" }],
});
FIG. 01 · Operating principles

A small atlas
of how Virouter routes.

Three rules govern every request that crosses the relay. They are boring on paper. They are why your bill stops surprising you.

  1. 01.

    One key, two labs.

    Virouter speaks /v1/chat/completions and /v1/messages. Swap a base URL, keep your SDK. OpenAI and Anthropic models behind a single Bearer token.

    → Keep your SDK, swap base URL, ship the same day.

  2. 02.

    Routed, never resold.

    We mark up tokens, not access. Your prompts and completions never enter a training set. Logs are scoped, retention is seven days, deletion is a button.

    → Clear retention windows. One-click deletion of the audit trail.

  3. 03.

    Failover by default.

    Each route has a primary and a degrade-tolerant alternate. When upstream coughs, we retry against the next healthy lane in < 90ms. You see one response.

    → Automatic retry to the next healthy lane. You see one response.

FIG. 02 · Billing manifesto

Tokens.
Not dollars.

Money is a side effect of language. Pricing in dollars per million tokens is honest math, but at the wallet level it is noise. Virouter shows you the only number that matters: how much language is left in the tank.

Console OSC-1 · Token telemetry
CH1 IN CH2 OUT

You think in prompts.
We meter in tokens.

A 50M token bag is roughly a small library of completed thoughts. A 1.2B token bag is roughly a season of agentic side projects. We won't pretend we can convert those for you in dollars — the dollar is too coarse a unit.

No tiers behind feature walls. No surprise overage. When a bag is empty, the relay tells you, then waits.

See the bags →
live · 50ms/div
  1. 1
    Buy a token bag. 50M / 300M / 1.2B tokens per month, sealed and stored in your wallet.
  2. 2
    Each request deducts input + output tokens. Upstream provider price is converted at registry rate.
  3. 3
    Overage at a fixed rate. $0.10/M on Spark, $0.07/M on Loom, $0.05/M on Forge — never a surprise multiplier.
  4. 4
    Usage visible in real time. Logs and per-key spend in the dashboard, exportable as JSON.
↳ Telemetry · live illustrative · sampled from production traffic
11:42:08.142ROUTEPOST /v1/chat/completions → claude-opus-4-8 · 1.4k in / 820 out186ms · sgp-1
11:42:08.310ROUTEPOST /v1/messages → gpt-5.5 (failover) · 220 in / 1.1k out241ms · iad-2
11:42:08.487DEGRADEopenai · transient 503 → retry on anthropic in 74msrecovered
FIG. 03 · Sealed bags of tokens

Three plans.
Three bags.

Each bag refills monthly. Unused tokens roll over for one cycle. Cancel anytime — the trunk closes the line at the end of the period, no autopay tricks.

Spark

50M tokens / mo

$5 /mo

For testing, hobby projects, and weekend agents.

  • OpenAI + Anthropic routes
  • OpenAI & Anthropic API surface
  • $0.10 per extra million
  • 7-day log retention
Start with Spark →
Best value

Loom

300M tokens / mo

$20 /mo

For active apps and agents in real production.

  • Everything in Spark
  • Priority routing lane
  • $0.07 per extra million
  • 30-day log retention & export
  • Multiple keys + scopes
Pick Loom →

Forge

1.2B tokens / mo

$50 /mo

For production-scale traffic and agentic workloads.

  • Everything in Loom
  • Dedicated channel
  • $0.05 per extra million
  • OpenAI/Anthropic audit log
  • Email + Discord priority
Scale with Forge →

Workload anchors · 50M ≈ a few side projects · 300M ≈ an active app with daily users · 1.2B ≈ a multi-agent product at scale.

A gateway should feel like an operator,
not a tax. — Virouter principle №3

FIG. 04 · Catalogue

Models on the line.

OpenAI and Anthropic only — focused routing, cleaner pricing, fewer moving parts for solo devs.

  • Best quality · Claude Opus 4.8, GPT-5.5
  • Best value · Claude Sonnet 4.6, GPT-5.4 Mini
  • Image · GPT Image 2
  • Long context · Claude Opus 4.x · 1M ctx
ModelProvider$/MTokContextTag
Claude Opus 4.8 Anthropic $5 / $25 1M ctx Priority
Claude Opus 4.7 Anthropic $5 / $25 1M ctx Opus
Claude Opus 4.6 Anthropic $5 / $30 1M ctx Opus
Claude Sonnet 4.6 Anthropic $3 / $15 1M ctx Balanced
Claude Haiku 4.5 Anthropic $1 / $5 200k ctx Fast
GPT-5.5 OpenAI $5 / $30 512k ctx Flagship
GPT-5.4 OpenAI $2.5 / $15 256k ctx General
GPT-5.4 Mini OpenAI $0.75 / $4.5 128k ctx Cheap
GPT-5.3 OpenAI $1.75 / $14 128k ctx Stable
GPT Image 2 OpenAI image vision Image

↳ See OpenAI + Anthropic routing docs →

Plug a key in.
Let it route.

Free to start. No card. 100k tokens on the house.

Create free account · 100k tokens → Read the docs