Models and tiers
Ichnus is powered by three models in the Krus, Mantus, and Corus family, named for the Ichnusa wind and the island of Curos. All three are available to every user. They differ in speed, context length, and reasoning depth, so the right choice depends on the task.
Model family
Krus
Krus is the fast, light model. It responds quickly and is well suited to short answers, drafts, summarization, and everyday tasks where speed matters more than depth. It has a 128,000-token context window, enough for a long document.
- Best for: quick drafts, Q&A, light editing.
- Trade-off: less depth on complex reasoning than the larger models.
- Available in: Ichnus and the API.
Mantus
Mantus is the balanced default and the model most people should reach for. It has a 256,000-token context window and a good ratio of depth to speed. If you are not sure which model to use, use Mantus.
- Best for: analysis, revision, planning, most professional work.
- Trade-off: on very hard reasoning tasks, Corus is stronger.
- Default: yes, Mantus is the default in new conversations.
Corus
Corus is the frontier model and the largest in the family, with a 1,000,000-token context window. It can work across an entire book-length document in one context. Corus is free to use within fair-use limits.
- Best for: long documents, multi-step research, complex reasoning, large codebases.
- Trade-off: slower responses, especially with long contexts, and limited by fair use.
- Fair use: free individual use up to a weekly limit; the current limit is shown on your account page.
When to use which
- Use Krus when you want a fast answer to a small question.
- Use Mantus for most work and as your default.
- Use Corus when you need the full 1,000,000-token context or the strongest reasoning.
The model selector in Ichnus shows a short description of each model so you can switch without leaving the conversation.
Fair-use limits on Corus
Corus is expensive to serve, so it is shared. Fair-use limits keep it usable for everyone and are measured per week, resetting on Monday. When a Corus request would exceed your remaining balance, the request is served by Mantus instead; nothing is lost and no additional request is needed.
- Limits apply to individual use of Corus, not to Krus or Mantus.
- Your current balance and reset time are shown on the account page.
- Research users who need more can apply for a program service allocation at cost.
- Fair-use limits are reviewed regularly and documented in the transparency report.
Context windows and tokens
The context window is the amount of text the model can consider in one request, counted in tokens. A token is roughly a fragment of a word; 1,000 tokens is close to 750 words of English. Attaching a file consumes context, so very long attachments leave less room for the response. For practical limits on the number of tokens you can send, see the API reference.
Choosing in the API
In the API you select a model with the model parameter, for example curos-ichnus-mantus-latest. See the API reference for exact identifiers and parameters.