Three-Axis Model

Omegon separates the operator's model route from provider-neutralmodel grade intent, thinking level, and context class. A provider-neutral intent can resolve according to declared policy. An exact concrete model override pins the selected route until /model unpin clears it. Selected and serving routes remain distinct when declared fallback compatibility is used.

Axis 1: Model grade intent

Grades are provider-neutral capability bands. They are not legacy slash aliases and they are not local endpoint names. Use /model grade <F|D|C|B|A|S> to express desired capability, or/model <provider:model> to pin an exact concrete route. Theexact grade policy means the requested grade must match exactly; it does not itself pin one concrete provider/model.

Grade	Use Case	Typical Examples
F/D	Cheap, fast, simple lookups and boilerplate	small hosted models, fast local models
C/B	Routine coding, execution, review, and general development	Sonnet/Codex-class routes, strong local models
A/S	Architecture, deep debugging, high-stakes reasoning	frontier reasoning and highest-grade provider routes

/model                   # Opens selector with available models
/model grade B           # Request a B-grade route, provider auto
/model provider local    # Restrict routing to local endpoints
/model policy minimum    # Allow requested grade or better
/model unpin             # Clear an exact concrete model override

Axis 2: Thinking Level

Controls extended reasoning budget — how much reasoning effort capable providers spend before responding.

Level	Use Case
off	No extended thinking. Fastest, cheapest.
minimal	Brief sanity check.
low	Light reasoning for straightforward tasks.
medium	Default. Balanced reasoning for general work.
high	Deep reasoning for architecture, debugging, multi-step problems.

/think high     # Manual override
/think off      # Disable for speed

Axis 3: Context Class

Controls context window capacity. Larger windows allow more conversation history and memory injection.

Class	Window	Use Case
Compact	128k tokens	Quick tasks, simple edits
Standard	272k tokens	Standard development work
Extended	400k tokens	Large refactors, multi-file analysis
Massive	1M+ tokens	Massive codebases, full-project comprehension

/context massive  # Pin to maximum context
/context compact  # Minimize for cost

Orthogonality

The axes are independent. You can request grade S + off + compact (frontier model, no extra thinking, small window) or grade D + high + massive (cheap model, deep reasoning, huge context). The route controller preserves those intents without pretending they are the same thing.

Simple file read → grade D + off
Architecture decision → grade S + high
Boilerplate generation → grade C + minimal
Large codebase audit → grade B + medium + massive

Provider routing

In the native TUI, /model route opens the model menu at its current route state,/model providers opens its provider view, and /auth status reports credential diagnostics. These TUI views do not list historical route leases and should not be read as a cross-surface command guarantee. Local endpoints are selected with /model provider local; they are not a legacy grade.

See Providers for the authentication and fallback boundaries.