switchboard
live model families
now routing · GPT-5.6

GPT-5.6 has three gears. Use all of them.

Switchboard sends demanding Codex work to Sol, everyday work to Terra, and the easy turns to Luna—automatically.

install & run— that's it.

You keep using claude code and codex like you always have.

switchboard·ROUTING·codex
── live feed ──
task
difficulty
model used
rename vars
●●●●●1
gpt-5.6-luna
refactor logic
●●●●●5
gpt-5.6-sol
run prettier
●●●●●1
gpt-5.6-luna
add JSDoc
●●●●●3
gpt-5.6-terra
design cache
●●●●●5
gpt-5.6-sol
fix typo
●●●●●1
gpt-5.6-luna
write tests
●●●●●3
gpt-5.6-terra
debug regression
●●●●●4
gpt-5.6-sol
the new model board

Every new model has a job.

These are the defaults. Every difficulty level remains configurable from Switchboard.

OpenAI · GPT-5.6new

One generation. Three gears.

Switchboard moves Codex between the new GPT-5.6 tiers instead of spending flagship capacity on every turn.

Sol
frontier
levels 4–5
Terra
balanced
level 3
Luna
efficient
levels 1–2
Anthropic · current Claudenew

Fable is back at the top.

The hardest Claude Code work can reach Fable, while Opus, Sonnet, and Haiku cover the work below it.

Fable 5
most capable
level 5
Opus 4.8
frontier
level 4
Sonnet 5
balanced
levels 2–3
Haiku 4.5
efficient
level 1
GPT-5.6 routing map

Sol when it matters.
Luna when it doesn’t.

Five difficulty levels move across the three GPT-5.6 tiers. Keep the flagship for complex work and let Terra or Luna take everything else. Every model and effort remains configurable.

5expert
→gpt-5.6-solxhigh
4hard
→gpt-5.6-solmedium
3normal
→gpt-5.6-terramedium
2easy
→gpt-5.6-lunamedium
1trivial
→gpt-5.6-lunalow
the result
Easy tasks go to cheap.
Hard tasks still get your best.
you stop
overpaying.
it costs
$1 for 1,000 requests.

Pay for what you use. No subscription necessary.

install & run— that's it.