NEW · Auto Mode

Stop picking models. The cheapest model, the best results.

An agent task makes dozens of model calls. Auto picks today's most popular, capable and affordable models in real time: it starts on the cheapest and upgrades only when necessary — spend less where you can, pay up only where you must.

Billed per routed model · No auto surcharge · Switch back anytime

auto — call log of one task
plan cheaper model $0.001
tool loop cheaper model $0.002
tool loop cheaper model $0.001
upgrade when needed stronger model auto-upgrade ⤴ $0.012
tool loop cheaper model $0.001
subagent cheaper model $0.001
compact cheaper model $0.001
··· 35 more calls ···
42 calls · $0.09 total all stronger models ≈ $0.45
Why Auto

Picking one model is picking a compromise

You can only pick one model — and without knowing the models, you're bound to overpay. Leave the worry to us.

Stronger models only

Great quality, painful bill. File reads, tool loops, context compression — all burning top-model rates.

Cheap model only

Complex planning tends to crash. One retry wastes more than tokens — it wastes your half hour.

Manual switching

You're not watching the task run. By the time you notice the crash, the rework cost is already sunk.

The core idea

One task. Every call gets what it deserves.

A typical refactor task: 42 model calls — only 5 truly need a stronger model.

Task timeline · 42 calls
Cheap-model base (37) Stronger-model upgrades (5)

Three strategies, three outcomes

All stronger models $0.45

Stable, most expensive

All cheap models $0.05

2 crashes + retries, +30 min rework

Auto Recommended $0.09

+$0.04 vs cheap models, stronger-model quality

Illustrative costs for the same task. Your actual bill may vary.

How it works

Economy-first: start cheap, upgrade only when needed

Auto doesn't bet on one "best model" — it picks today's most popular, capable and affordable models in real time, putting every call on the right one.

Base tier: Cheaper models

90% of calls — tool loops, subagents, compaction — run on fast, affordable models.

Upgrade only when needed

Consecutive failures and retries in one task, or planning that clearly strains the floor — only then does Auto move to the stronger tier; one success sends it back.

Safety net: stronger model falls back to base

If the upgraded stronger model is temporarily unavailable, Auto drops back to the base and keeps going — no interruption, no failed task. A new task retries the stronger model automatically.

Your model pool, two tiers

Auto is exclusive to OpenClacky AI Keys — one key, every model.

Role Models When
Base (always on) Cheaper models Tool loops / subagents / compaction
Upgrade tier (on demand) Stronger models Complex planning / retries
Transparency

Every route visible. Every cent accounted.

Auto never does anything behind your back.

Live in your status bar

See which model each call actually routed to, and what it cost, in real time.

This month · by model

Cheaper models — 1,162 calls · $1.95
stronger model — 122 calls · $1.25

Total this month: $3.20

FAQ

Will Auto secretly route me to expensive models?
No. Every route shows in your status bar in real time; billing follows the actual model used, and the billing page summarizes spend by model and by day.
Does Auto always use the same base model?
No. The base dynamically selects a capable, cost-effective model; only failure signals trigger a temporary upgrade to a stronger model.
Can I turn Auto off and pick models myself?
Anytime. Auto is just the recommended default — every model stays one click away in the model switcher.
Does Auto cost extra?
No. Auto is a routing capability. You're billed for the models actually used — no surcharge.
What do I need to do when a new model is released?
Nothing. Auto keeps up with new releases — whenever a newer model is better or cheaper, it's live for you.

Make model selection a thing you never do

Top up once — one key, every model, routed on demand.

Get credits

Already have an account? Sign in