ek-dev- key, a curated :dev model list, and unlimited tokens with no credit burn. It does not replace a subscription or pay-as-you-go balance; it stacks with them.
Tiers
Token usage is unlimited on both tiers. Past full-speed headroom you stay online in a slower lane (see Limits) — you are never cut off for burning tokens.
Models
Only:dev models. Only with an ek-dev- key. Use your master key for everything else in the catalog.
All of them support tools on
/v1/chat/completions, /v1/messages, and /v1/responses.
Context: the three long-window models above are 524K on Turbo and 262K on Lite. That limit is enforced on every request. GET /v1/models is public and shows the Turbo max; send your ek-dev- key (list) or call authenticated /v1/models/{id} to see your tier’s window.
Series weight (0.25× / 1× / 3×): raw tokens are multiplied when scoring soft headroom (ceil(raw × weight)). Flash burns headroom slower; Heavy burns it faster. Dashboard per-model charts stay unweighted.
Quickstart
- Subscribe at Dashboard → Console → Coding Plan.
- Copy the
ek-dev-...key from that tab (regen invalidates the old key immediately). - Call the API:
Limits
These controls keep the pool usable. None of them are a hard token cap.Concurrent slots
A request holds its slots until it finishes. If it needs more slots than are free →
429 with Retry-After.
In the slow lane, concurrency drops to 1 on Lite and 2 on Turbo. Heavy needs 3 slots, so it is refused with 403 / model_slots_exceeded until the healthy cap returns. Flash and Standard still run. On Lite, Heavy is never available — upgrade to Turbo for glm-5.3:dev.
Full-speed headroom
Past either budget you keep running, but admission slows: concurrency drops (Lite → 1, Turbo → 2), RPM shrinks, and inter-request spacing grows with consumption (up to ~30s when far over). The dashboard meters show remaining headroom. Series weight applies to these budgets; see Models.
Requests per minute
Continuous refill: 8/min Lite (burst 4), 20/min Turbo (burst 8). Every start counts once, including Heavy. Slow lane: 3/min (burst 2). Over the limit →429 with Retry-After. Headers: X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset. These RPM 429s do not count toward the concurrency pause below.
Concurrency pause
About 100 concurrency429s in a usage day can pause admission for 5 hours. Paused requests return 403. The key remains valid; service resumes at the dashboard time. Regenerating the key clears an automatic pause.
Low-interactivity mode
If usage looks far beyond one person coding, the seat may enter this mode for ~24 hours:- Light load: unchanged.
- Busy periods: may queue behind interactive traffic; retry if needed.
- Models, quality, and entitlement stay the same. Not a suspension.
Fair use
One person, one seat. Lite and Turbo are alternatives, not stackable — upgrade instead of buying twice. You may use the key on machines you personally use (home, work, laptop, CI, VPN). All of that counts as the same seat. Suspended for: second seats under identities you control; sharing the key; pooling seats for throughput; resale or access-as-a-service; backends that serve other people; unattended multi-user production traffic. Chargebacks are account-level. High-duty-cycle or multi-user traffic belongs on a credit-billed master key.Errors
Billing
- Monthly via Creem, auto-renew, non-refundable. Cancel anytime; the key stays live until period end.
- Founding / beta buyers keep their launch price while the subscription stays active. Cancel and resubscribe later → current public price.
- Upgrade Lite → Turbo in the Creem portal; the same
ek-dev-key picks up Turbo limits.
FAQ
Is usage really unlimited?
Is usage really unlimited?
Yes — no hard token stop and no per-token charge. Soft headroom only slows concurrency, RPM, and spacing when you are far past typical interactive use.
Can I hold two seats?
Can I hold two seats?
No. One person, one seat. Upgrade to Turbo for more parallel slots.
What happens when the subscription ends?
What happens when the subscription ends?
The
ek-dev- key is deactivated. Resubscribe to restore it.Why can’t Lite run glm-5.3:dev?
Why can’t Lite run glm-5.3:dev?
Heavy uses 3 concurrent slots. Lite has 2, so admission refuses it with
model_slots_exceeded. Upgrade to Turbo (6 slots) to run Heavy.