Introducing OptimAI Router: one network, two ways in
Open models have never been better. You can now pick from dozens of capable model families for chat, coding, writing and reasoning, and new ones arrive every month.
Using them is another story. Running a model yourself means hardware, setup and upkeep. Renting one through a cloud means someone else's prices and someone else's rules. Meanwhile, millions of capable computers sit idle every night.
OptimAI Router connects those two sides. Builders get open models through one API. People with capable computers can serve those models and earn for the work.
Why we built Router
Every product starts with a bet. Ours came from watching three trends move faster than almost anyone expected.
1. Open models became good enough
Two years ago, the best open models were clearly a step behind the closed frontier. That gap has nearly closed. Stanford's 2025 AI Index found that on the Chatbot Arena leaderboard, the gap between the best closed and best open model shrank from 8% in January 2024 to 1.7% by February 2025 [1].
They also became small enough to run at home. Epoch AI found that open models able to run on a single gaming GPU match the frontier from roughly 6 to 12 months earlier [2]. In August 2025, OpenAI released its first open-weight models since GPT-2, and the smaller one, gpt-oss-20b, runs on devices with 16 GB of memory under an Apache 2.0 license [3]. A model that answers well on a desktop computer is no longer a demo. It's ordinary.
2. Demand keeps growing as AI gets cheaper
Cheaper, better models don't shrink the market. They expand it. After the DeepSeek R1 release in early 2025, Microsoft CEO Satya Nadella pointed to the Jevons paradox: as AI becomes more efficient and accessible, people use far more of it [4].
The numbers agree. MarketsandMarkets estimates the AI inference market at about USD 106 billion in 2025, growing to about USD 255 billion by 2030 [5]. Usage of open models is real, not theoretical. An OpenRouter and a16z study of more than 100 trillion tokens found that open-weight models handled about one third of all usage by late 2025 [6].
3. The industry turned toward open
The largest AI companies now treat open models as part of the future, not a side project. In January 2025, Sam Altman said he believed OpenAI had been on the wrong side of history on open source and needed a new strategy [7]. Months later, OpenAI shipped gpt-oss [3]. In July 2026, NVIDIA's Jensen Huang used his first-ever post on X to share an open letter on open models:
"The world needs both frontier closed models and frontier open models." [8]
The gap we saw
Put those together and a gap appears. Open models are good, small and widely used, yet serving them still mostly means renting capacity from large data centers. Meanwhile, capable computers that can run these models sit idle for most of the day.
We asked a simple question: what if the people who own those computers could serve open models to the builders who need them, through one trusted network?
Why OptimAI is the right place to build it
OptimAI already lives on people's computers. Core Node is where our community runs the network, and OptimAI accounts, Credits and Points already connect that community. Router extends what our community already does, instead of starting from zero. Providers get a guided way to serve models from a computer they already run. Builders get one API, backed by that community.
We also kept the hard parts honest from day one. Serving is opt-in. Providers set their own prices. Builders set their own limits. Charges are based on confirmed usage.
Why x402
AI is starting to pay for itself. Agents and apps increasingly need to buy compute, data and services on their own, in small amounts, without a human filling in a card form each time. Card networks weren't designed for that. Stablecoin payments over the web are.
x402 is the open standard built for this. Coinbase introduced it in May 2025 [9], launched the x402 Foundation with Cloudflare [10], and Google brought x402 into its Agent Payments Protocol as a stablecoin extension [11]. By August 2026, Coinbase reported about 205 million x402 transactions, worth roughly USD 53 million in total [12].
OptimAI already supports x402 for adding credits. The bigger ambition comes next: x402 pay-as-you-go, where an app or agent pays for each request as it goes (Coming soon). Open models, served by people, paid for by software. That's the network we want to help build.
For builders: open models, one API
Router gives you a single place to discover open models, try them and call them from your app.
- One connection. Router exposes a familiar chat completions API, with streaming and non-streaming responses. If you've used an OpenAI-style client, you'll feel at home. Router supports a focused subset of that API, centered on text chat, and the docs list exactly what's covered.
- You choose the model. Every request names a specific model. Router never quietly swaps in a different one. If no provider can serve your choice right now, you get a clear answer instead of a surprise.
- Try before you build. The Playground lets you send real requests from your browser before you write a line of code.
- Keys and limits you control. Create API keys under your OptimAI account and turn on paid use for each one. Your account has a monthly spending cap, you can give a key a lower limit, and every request has a maximum charge. Usage history shows what each key did.
- Simple top-ups. Requests are paid with OptimAI Credits, and you can add credits with x402 payments. You're only charged for usage that's confirmed.
For providers: your computer can serve
If you have a capable computer, Router lets it do useful work for builders around the world, on your terms.
Setup happens in OptimAI Core Node, in four steps:
- Computer. Router checks what your machine can run.
- Model. Choose a model that fits your hardware. Router runs a short local test to make sure it answers before you go further.
- Pricing. Router suggests a starting price. Keep it or set your own.
- Serving. When you're ready, press Start serving. Stop whenever you like.
When your computer completes requests, you earn credits. While you're serving, you also collect Points. Your Home screen shows Points, Credits earned and Requests completed side by side.
Serving is always opt-in. Your settings stay yours, and serving starts only when you choose to start it.
How a request travels
A builder's app sends a request to Router and names the model it wants. Router looks for a provider that serves that model and fits the builder's limits. The request is assigned to that provider. The provider's computer generates the answer and streams it back through Router to the app. Usage is then confirmed and settled in credits: the builder is charged and the provider earns.
Good to know
We'd rather you trust Router than be surprised by it, so a few honest notes:
- Providers' computers process the requests they serve. Don't send secrets or sensitive personal data through Router.
- Router currently serves text chat. Some model families can do more (vision, embeddings, audio), but those capabilities aren't available through Router yet.
- Credits stay in the OptimAI ecosystem. Withdrawals and token payouts aren't available today.
What's next
Router is starting with the essentials: a dependable way to call open models and a fair way to serve them. Next on the list are more ways to connect and pay, including MCP and x402 pay-as-you-go for individual requests (both Coming soon), along with broader model support as providers bring more capacity online.
Get started
- Building something? Open the Router dashboard and get an API key.
- Have a capable computer? Set up Router in Core Node and start serving.
Learn more at router.optimai.network.
A more open tomorrow starts with what you build today.
Sources
- Stanford HAI, The 2025 AI Index Report (chapter 1 PDF)
- Epoch AI, Frontier AI capabilities can be run at home within a year or less
- OpenAI, Introducing gpt-oss (August 5, 2025)
- Satya Nadella on X, reported by GeekWire
- MarketsandMarkets, AI Inference Market worth USD 254.98 billion by 2030 (see also Grand View Research)
- OpenRouter and a16z, State of AI: An Empirical 100 Trillion Token Study (also on a16z.com)
- TechCrunch, Sam Altman believes OpenAI has been on the wrong side of history concerning open source (January 31, 2025)
- Fortune, Jensen Huang's first X post on open models (July 24, 2026)
- Coinbase, Introducing x402: a new standard for internet-native payments
- Coinbase, Coinbase and Cloudflare will launch the x402 Foundation
- Google Cloud, Announcing Agent Payments Protocol (AP2), and Coinbase, Google AP2 + x402
- AI agents have processed 205 million transactions via x402, Coinbase says (August 2026)
