As Nexus prepares for launch, one of the most exciting capabilities it will offer from day one is seamless access to large language models hosted on distributed infrastructure: powered by our ongoing collaboration with io.net.

Thanks to our partnership with io.net, Nexus will support a range of pre-hosted decentralized language models, including powerful options like Qwen-235B and DeepSeek R1, running on io.net’s distributed compute layer. As soon as Nexus goes live, users can select these models directly from the setup wizard and start building and testing AI agents in minutes.

What started as a shared vision between our teams has now evolved into a working integration that’s ready to use, bringing together io.net’s hosted inference layer with Nexus’s modular AI framework in a way that makes both platforms more accessible, and more useful.

From day one, users will be able to choose io.net-hosted LLMs, like DeepSeek R1 or Qwen-235B, as the model powering their AI agents. This marks an important step forward in making Nexus truly open, decentralized, and accessible.

What This Integration Enables

Thanks to this partnership, every Nexus user will be able to:

• Use decentralized hosted LLMs from io.net’s global compute network, with zero infrastructure setup.

• Choose from pre-hosted models, including powerful open alternatives like Qwen-235B and DeepSeek R1.

• Start for free using a personal io.net API key (daily usage limits apply, but they’re generous enough to fully explore Nexus capabilities).

• Get started instantly via our Setup Wizard: just connect your wallet, log in, and select the io.net option to use the preconfigured Qwen-235B model. No payment or server config will be required.

• To uncap your API usage and take full advantage of the power of open-source models through IO Intelligence, reach out to partners@io.net

io.net offers free LLM usage with generous daily limits.

For developers, this means you can start building autonomous AI agents with real capabilities, running on decentralized infrastructure, without touching an OpenAI key or deploying your own inference server.

For the broader ecosystem, it’s a glimpse of what permissionless AI infrastructure looks like in action.

A Partnership That Works Both Ways

Through this integration we are creating a smoother path for new users to try Nexus, while introducing them to the power and promise of io.net's decentralized inference layer: for Nexus users, io.net offers a no-cost, no-hassle way to start using large models right away and for io.net, Nexus serves as a natural entry point for developers and agents looking to put hosted models to use in structured, interactive ways.

It’s a great example of mutual value: two platforms reinforcing each other’s mission consisting of decentralization, openness, and usability, by working together where it matters most: the user experience.

A Glimpse of What’s Possible

Some of the immediate use cases unlocked by this integration:

• Agent experimentation without a credit card: Test autonomous workflows using io.net’s hosted Qwen-235B model, integrated directly in the Nexus setup.

• Multi-model agent orchestration: Easily switch between OpenAI, io.net, or local models depending on your needs.

• Deploy custom agents for search, data transformation, API interaction, or DAO tooling without worrying about hosting the model backend.

This is just the start. Over time, we envision deeper links between Nexus and io.net. As both ecosystems evolve, we see a shared opportunity to redefine how agents interact with compute, and how developers build intelligent systems that are open, efficient, and user-controlled from end to end.

About io.net

io.net is a decentralized distributed compute network that lets ML engineers deploy large-scale GPU clusters in seconds, saving up to 90% compared to monolithic Big Tech compute providers. Our infrastructure seamlessly aggregates resources from multiple locations and deploys them into a single cluster at massive scale. io.net has successfully supported training, fine-tuning, and inference for a wide range of ML models, including Llama-4-Maverick-17B-128E-Instruct-FP8, QwQ-32B, and DeepSeek-R1.