Introducing Infinity: Model and Harness Built for Traders

The most common assumption about AI is that intelligence is what matters most. The smarter the model, the better the answers. For trading, we think that assumption is wrong, or at least incomplete.

A good AI is not the one with the most raw intelligence. It's the one with access to the best tools. Ask the smartest general-purpose LLM why a token sold off this morning and it will reason carefully from information that is months old.

Markets make this unusually stark. Trading is more about information than pure intelligence. Edge comes from seeing the unlock schedule, the filing, or the flow before it gets priced in. A model that cannot see those things can only describe markets in general terms, which is why most AI trading products feel like the same product: familiar lagging indicators wrapped in a chat interface, with the conversation standing in for insight.

We built Infinity from the opposite starting point. Instead of asking how smart we could make a model, we asked what a model would need to see to be genuinely useful to a trader, and then we built the retrieval and orchestration infrastructure to let it see all of it.

A model that understands trading, not just charts

Infinity pairs a frontier-class reasoning model, post-trained on publicly available on-chain trade data from Hyperliquid, with a purpose-built agentic harness that connects to 16 live data sources, including Coinglass, CoinGecko, and Nansen, and fuses them into a single normalized research view. Real-time macro and news sentiment streams sit alongside derivatives, on-chain, and fundamental data in the same context.
When you ask Infinity a question, a planner model running on our own GPU infrastructure performs query decomposition in real time: it classifies the question, decides which data sources it requires, and dispatches structured tool calls in parallel rather than dumping every feed into context. The retrieved results are grounded with citations you can verify yourself, so every claim traces back to a source rather than to the model's priors.

Under the hood, data arrives through two latency tiers. Anything that moves tick by tick, such as funding rates, open interest, liquidations, and exchange inflows and outflows, is fetched live at query time through direct API connections. Slower-moving unstructured data, such as news, filings, and research, flows through a streaming ingestion pipeline that indexes fresh documents every hour.

The result is behavior that feels categorically different from a chatbot. Ask Infinity about a token and it checks the unlock schedule. Ask about an earnings move and it reads the filing. It works less like a chatbot with opinions and more like an analyst with a live terminal, because structurally that is what it is: a capable reasoning engine wired into the data traders actually depend on.

This is also why Infinity improves in a way that a frozen model cannot. Intelligence improves in step changes, but tooling compounds. Every new data source, every refinement to the planner's routing accuracy, and every hour of fresh ingestion widens the gap between what Infinity can see and what a general-purpose model can. The model is a component. The harness is the product.

We look forward to user feedback as we continue to improve Infinity.

Research preview

Infinity is live today in research preview, and it is free until 27 August 2026 12:00 UTC.

You can access Infinity a few ways:

- Mobile iOS /Android: Tap "Research" on the home page.
- Web Terminal : Click "Ask Infinity" in the navigation bar at the top right.

Based AI: Infinity is also available through ai.based.one, our unified AI platform, alongside a growing set of models for coding and general reasoning, including Kimi K3, Qwen, GPT-5.6 Sol, and Claude Fable 5.

Infinity is in the research preview. Outputs may contain errors and should not be considered financial advice. Free access ends 27 August 2026.