Is it cheap right now?

Live peak / off-peak windows for the LLM providers that actually have them. Times are shown in your local timezone — verify the detection below.

Right now, in your timezone

—

Next change —

Next window where all three are cheap —

Today in your local time

cheap / off-peak peak / full price

Timezone we detected for you

Everything above is computed from this. If it's wrong, the windows will be wrong — pick the right one.

Override:

Which product does each policy actually apply to?

DeepSeek — pay-as-you-go API only (api.deepseek.com, both OpenAI- and Anthropic-format endpoints). DeepSeek has never shipped a coding plan or subscription, so there is no other surface it could apply to. Concurrency limits (2,500 flash / 500 v4-pro) are flat and not time-based.
Z.ai GLM — the GLM Coding Plan subscription only. Z.ai's pay-as-you-go API has flat per-1M pricing with no peak column at all. The 50% off-peak credit rate is plan-wide, not ZCode-specific — Z.ai's FAQ states all supported tools share the same usage quota, so Claude Code and Cline users get the identical discount. Legacy V2 plans are worse at peak: GLM-5.3 burns 3×, Flash 1.2×.
ZCode-only extras are capacity-based, not clock-based. ZCode users additionally get idle-time tasks — the entire unattended run is free, consuming no plan quota at all, dispatched from a global queue when spare capacity exists with no promised start time. It also auto-issues quota reset cards during off-peak, restoring your 5-hour quota to 100%. Both are strictly better than the 50% credit rate, but neither has a published window: Z.ai states the reset-card off-peak window is not the same as the idle-time-task window, and both are dynamic — "go by what the app shows." The widely-repeated 150% quota boost expired 2026-08-31; it was a limited-time promo, not a standing benefit.
Alibaba / Qwen — the Token Plan (China region) night credit discount. Alibaba runs three different night policies on the same 22:00–08:00 UTC+8 window: Token Plan credits (60% off qwen3.8-max/flash, 50% off DeepSeek models), Model Studio pay-as-you-go (per-model, limited-time), and the separate "Night Plan" SKU (0.2×/0.4×, Pro/Max tiers, Qoder CN + Meoo web only — not Meoo CLI or the plain API). The international Coding Plan is request-based and has no time-of-day discount.

Treat today as a Chinese public holiday — DeepSeek excludes them from peak entirely, but publishes no holiday calendar, so we can't detect it. Tick this and DeepSeek reads all-day off-peak.

Providers deliberately not shown here

Claude — had a peak-hours limit reduction until Anthropic removed it on 2026-05-06. Its 5-hour session limit still exists but is usage-keyed: it starts on your first message, not at a clock boundary, so no countdown is possible.
OpenAI Codex — 5-hour + weekly windows, same usage-keyed problem. Its Fast 2.5× / Ultrafast 8× are speed multipliers, not peak pricing.
Gemini — Batch API is 50% off, but that's a latency choice (async, up to 24h), not a time window.
Cline — no policy of its own and can't have one; it inherits whatever provider you've connected.
Kimi, MiniMax — no published peak window. MiniMax states its peak traffic hours are dynamically adjusted based on cluster load, so any countdown would be invented.