Meet asi1-ultra: Your AI's Deep-Thinking Mode
Sometimes you need a quick answer. Sometimes you need your AI to read 800 pages, weigh tradeoffs, and produce a real recommendation. asi1-ultra is the model for the second one.
Most of the time, you do not need a model that thinks for thirty seconds before it answers. You want a quick reply, a tool call, a summary - and you want it now.
But every once in a while, the question is bigger than that. You want to audit a contract. You want a research agent to read fifty sources and write you the truth. You want a coding agent to fix your repo, run the tests, and keep working until everything is green.
That is the work asi1-ultra is built for.
What asi1-ultra Is
asi1-ultra is the deepest-reasoning model in the ASI:One family. It is the same product, the same Personal AI, the same memory and tools - just with more thinking power and a much bigger budget for getting work done in a single turn.
The headline numbers, in the language that matters:
- 200,000 tokens of context Drop in entire books, codebases, contracts, or research libraries. asi1-ultra reads it all in one go.
- Up to 500 tool calls per turn Long agentic flows that used to require chaining requests now run end-to-end in a single conversation. Search, read, evaluate, search again, write - asi1-ultra sustains it.
- Deepest reasoning in the family Multi-step analysis with intermediate validation. Built for catching the subtle issues, not skimming.
- Same Personal AI you already use Your memory, your skills, your connected tools - all of it. asi1-ultra is a depth upgrade, not a different product.
Where asi1-ultra Shines
Some workloads are bounded. A weather check. A calendar lookup. A quick summary. The default model handles those beautifully and asi1-ultra would be overkill.
asi1-ultra is for the unbounded ones. The tasks where the answer depends on connecting evidence, exploring branches, and not giving up at step three.
- Deep research "Find every credible source on this topic from the last six months, weigh the evidence, and tell me what the consensus is and where it breaks down." That is fifty tool calls of work. asi1-ultra does it in one turn.
- Long agentic runs A travel-planning task that has to coordinate flights, hotels, restaurants, and a calendar across three days. A code-review task that has to read every PR in a repo. A contract audit that has to cross-reference clauses against every other clause. asi1-ultra is built to sustain these.
- Code reviews and audits When you want the model to actually read the code, not just glance at it. asi1-ultra reasons across larger codebases and catches the things a faster model would miss.
- Strategic planning Multi-constraint problems where the right answer depends on understanding tradeoffs, not just retrieving facts. Hiring plans, product roadmaps, competitive analyses.
- Document analysis at scale Contracts, regulatory filings, legal briefs, scientific papers. Anything where the depth of reading matters.
The Tool Budget Is the Big Deal
500 tool calls per turn sounds like a developer spec until you think about what it means in practice.
Every time your AI searches the web, that is a tool call. Every time it reads a document, queries an agent, posts to a channel, checks a calendar, or runs code - tool call. The more tool calls a model can chain together in one turn, the bigger the task it can actually finish without forcing you to babysit it.
With asi1-ultra, your AI can plan a complex task, execute it end-to-end, and deliver the result in a single response. No "let me do the next step" handoffs. No state to manage between turns. Just the work, done.
Think of it this way
How to Use It
On your Personal AI, asi1-ultra is available as a model selection on paid plans. You can pick it for a single conversation or set it as your default - your AI keeps all of its memory, skills, and connections.
For developers, asi1-ultra is live in the API today. Same key. Same request format. Same OpenAI SDK compatibility. Just change the model field:
Switching models is one line
"model": "asi1" to "model": "asi1-ultra" and you are done. Same context window. Same tools. More depth.See the asi1-ultra developer docs for specifications, code examples, and a migration walkthrough.
When Not to Use It
asi1-ultra is the right pick when you want depth. It is the wrong pick when you want speed. If your task is bounded, fast, and well-defined - a triage decision, an autocomplete suggestion, a real-time chat reply - the default asi1 model (or asi1-mini if speed dominates) is the better choice.
You will not break anything by using asi1-ultra for everything. But you will pay for thinking you do not need. Reach for it when the task warrants it.
For a side-by-side comparison and a decision tree, see Model Selection.
What This Unlocks
For the first time, asking your Personal AI to "research this properly," "audit this contract," or "fix this whole repo" is not an aspiration. It is a button you can press.
That is the unlock. asi1-ultra is the model for the work that used to be too big.
Keep Reading
Ready to go deeper?
Switch your Personal AI to asi1-ultra and ask it the question you have been holding back. Or pick up the developer docs and ship a deeper agent today.