ArcRouter

93/100 for arcrouter. agents.circle.com/sell/score…

ArcRouter is now part of the @nvidia Developer Program. We're building the router for agents: one OpenAI-compatible endpoint for every model, paid per call in USDC on Arc. No account, no API key, no card. Membership gives us NVIDIA's SDKs, NIM microservices and early access to new tools, and puts us closer to the open Nemotron models. More models, served better, for agents that pay their own way.

OpenAI's (@OpenAI) new dots, the always-on agents in ChatGPT, run on GPT-6 Astra. That same model is now on ArcRouter, so your own agent can use it too. No ChatGPT Pro plan, no OpenAI account, no API key to babysit. Your agent pays per call in USDC on @arc and only for the tokens it uses. And it's 50% off for launch: $5 in / $25 out per 1M tokens. model="gpt-6-astra" model="gpt-6-astra-pro" when you want it to think longer Same OpenAI-compatible endpoint, so it's a one-line change.

ArcRouter is now an MCP server. Every model in our catalog, available as tools inside your agent. No package, no SDK, nothing to install. Works with Claude Code, Cursor, and any MCP client. What your agent gets: → ask: send a prompt to any model, from Claude to DeepSeek to private TEE models → compare: the same prompt to up to 4 models in parallel, answers side by side → list_models: the full catalog with live prices → balance: what it has spent, down to the token Every answer tells you what it cost. A second opinion from a smaller model often costs a fraction of a cent. And it spends from a prepaid key, so your agent can never spend more than you put in. New wallets get 5M free tokens to try it. Setup guide at docs.arcrouter.co/mcp

AI agents can write code, trade, run research, and ship products. They still can't sign up for an API key. Every API on the internet was built for a human. Create an account. Verify your email. Add a credit card. Prove you're not a robot. Paste the key into .env. An agent fails at every step. And it doesn't stop at one. Every model provider means another account, another key, another monthly bill. One leaked key keeps spending until someone notices. An onchain router flips the model: → Wallet = identity. No signup. The address is the account. → Balance = budget. An agent can only spend what its wallet holds. → Payment = receipt. Every paid call settles in USDC on Arc, auditable by anyone. → Machine speed. Request, get the price, sign, get the answer. No gas, no human in the loop. One wallet. One endpoint. Every model. The next wave of API calls won't come from people clicking buttons. It'll come from agents with budgets. They need infrastructure that speaks money natively. Give your agent a wallet, not a password.

The onchain inference router for A̶r̶t̶i̶f̶i̶c̶i̶a̶l̶ Super Intelligence on @arc.

Trust is essential, but agents also need access to intelligence. That's where ArcRouter comes in. Pay-per-use AI inference, settled on @arc with USDC. Every request has a cost. Every payment is verifiable. Building the intelligence layer for the agentic economy.

1,033,938 $ROUTER burned. Our first 1M+ burn. Buyback #3: 140 USDC swapped for $ROUTER, sent straight to 0x…dEaD on Arc. Both txs are public. Total burned so far: 1,880,497 $ROUTER across 3 buybacks. Ledger: arcrouter.co/router

ArcRouter now speaks 中文 🇨🇳 The site, the console and the docs are all available in Simplified Chinese: → arcrouter.co/zh → docs.arcrouter.co/zh One click on 中文 in the nav, or switch back to EN anytime. ArcRouter 现已支持简体中文：官网、控制台与文档全面上线。

ArcRouter docs now have their own home: docs.arcrouter.co Everything you need to ship on one focused site: → Quickstart: first call in a minute → Pay per call with x402, no key, no account → Prepaid keys, caching and refunds, errors Swap one base URL and your OpenAI SDK just works.

Arc is being built for an economy where AI agents can transact autonomously. ArcRouter brings the intelligence layer to that economy: agents can access inference on demand and pay per use in USDC directly on @arc. No subscriptions. No prepaid credits. Just compute when the agent needs it.

20M+ tokens through ArcRouter in its first week. 989 calls, every one metered per token and settled in USDC on @arc. One OpenAI-compatible endpoint, any model.

5M free tokens. Every model. New to ArcRouter? Sign in with your wallet and get 5,000,000 tokens on the house: Claude Opus 5.5, GPT-5.5, Grok 4.7, DeepSeek, Kimi and the rest. No card. No deposit. One signature.

Repeat a request, pay half. ArcRouter has a response cache. Identical calls replay instantly from cache and bill at 50% off, so you can cut inference costs by up to 50%. On by default at temperature 0. One header for everything else.

BlackRock is describing the future of machine-native payments. We're already building for it. ArcRouter gives AI agents pay-per-use inference, settled in USDC on @arc. No subscriptions. Just compute when agents need it.

Grok 4.7 is live on ArcRouter. xAI's flagship for coding and agents, on the same OpenAI-compatible endpoint as Claude, GPT-6, DeepSeek and the rest. Paid per call in USDC on @arc. No account, no card. model="grok-4.7"

Onchain AI narratives are heating up.