How Athena works
Athena is aixplain's model router: one model you call that picks, for every request, which model should answer it. This article explains how Athena decides, where you can use it and what you pay.
What is Athena?
Most of the steps an agent takes are simple: reading a file, renaming a variable, answering a quick question. A few need real reasoning, like finding the cause of a bug or planning a change across many files. Sending every step to the strongest model means paying frontier prices for routine work.
Athena reads each request and sends it to the kind of model it needs. Hard reasoning goes to a frontier model, and routine steps go to faster models that cost a fraction as much. You call Athena like any other model, and it makes the choice for you.
Key capabilities
- One model to call: Athena has one Marketplace ID, so you use it wherever you can use a model.
- A choice per request: each new request is judged on its own, so one session can use all three tiers.
- Frontier quality where it counts: when Athena isn't sure about a request, it sends it to the strongest tier.
- Fallback: when a model isn't available, Athena falls back to another one, so the work keeps going.
- See every choice: every reply records the model that answered it.
How Athena decides
For each request, Athena sorts the work into one of three tiers:
| Tier | What goes there | Name in the API |
|---|---|---|
| Deep | Hard design and reasoning, and multi-step planning, on a frontier model. | fat |
| Standard | Routine work: reading files, a rename, a docstring, running a command. | slim |
| Quick | Conversation: a question, an explanation, a summary. | fast |
A fast classifier makes the choice, and it adds about a tenth of a second to each call. The request is then served by the model behind that tier. aixplain chooses and maintains the models behind Athena's tiers.
A task is judged once, and the tool steps that follow keep its tier. They move up to a stronger model only when a tool result shows an error.
Route between your own models
To keep the routing but choose the models behind each tier yourself, use Automode. See Set up Automode with your own models.
Where you can use Athena
- In Koder and Omni: choose Auto in the model picker. See Use Athena in Koder and Omni.
- In your agents: set Athena as the agent's LLM. See Run your agents on Athena.
- In your own apps and AI projects: call it with an aixplain API key, through the OpenAI-compatible API or through MCP. See Call Athena from your own apps.
Pricing and usage
Athena has no price of its own. Each request costs what the model that answered it costs, and the charge comes from your aixplain credits like any other model. For plans and credits, see Pricing.
Privacy and data
Requests go to aixplain and to the model that serves them. Check a model's data retention on its Marketplace card.
Limitations
- Athena doesn't beat the strongest single model on quality. If every solve matters more than cost, pick that model directly.
- You can't choose the models behind Athena's tiers. Use Automode for that.
- Savings depend on your mix of requests. Work that is mostly hard reasoning saves the least.
Frequently asked questions
Is Athena the same as Automode?
Athena runs on Automode. Athena is the version aixplain manages, with models we choose, and you can call it anywhere. Automode lets you choose the models yourself. See Set up Automode with your own models.
How do I know which model answered?
Every API response carries a router field with the tier and the model that answered. See Call Athena from your own apps. If you'd rather use one model for everything, choose it in the model picker or name it in your code.
Does Athena train on my data?
No. Athena sorts requests with a classifier that needs no training on your data.
Get help
- Report a problem: email care@aixplain.com with what you asked, the model you used and when.
- See what changed: read the release notes.
Still need help?
Report a bug on GitHub or write to us, and include your version and operating system.