Akili Fundi & Akili Core
Two models, one context window size, switchable mid-conversation. Most work starts on Core and steps up to Fundi when the thinking gets hard.
| Model | Reach for it when | Context |
|---|---|---|
| Akili Fundi (Code) | The work has stages — long agent runs, refactors, migrations, anything that has to hold a plan together across many steps. | 1,000,000 tokens |
| Akili Core | Chat, search, quick edits, reading and summarising. Fast and inexpensive; the right default. | 1,000,000 tokens |
Switching
Use the model dropdown in the chat bar. The switch takes effect on the next turn, and the context gauge and cost meter re-scale to whichever is active — so you can start a session on Fundi, let it plan and build, then drop to Core for the cleanup.
From the CLI, /model lists what is available and takes the name you were just shown:
/model
model: Akili Fundi (Code)
Akili Fundi (Code), Akili Core
/model Akili CoreWhat a 1M window actually buys you
A large window is not permission to send everything. Akili Code builds a context pack: it seeds from your question, spreads relevance across the dependency graph, and sends the files that matter. The window means long agent runs do not fall over halfway; it does not mean dumping your repository into every request.
The gauge in the chat bar shows real usage against the real window. When a conversation approaches it, Akili Code compacts the history rather than truncating it silently.
Cost
The meter reads actual usage from the API's own response — not an estimate. Cache hits are credited, so a repeat turn over the same context bills at a fraction of the first. Set per-conversation and per-day caps in Settings; a run that crosses one is stopped mid-flight, not reported afterwards.
The cheapest single change: prefer targeted edits over whole-file rewrites. edit_file sends a snippet diff; rewriting a large file re-sends all of it on every revision.