01 — THE ONLY CLOCKPeak or off-peak, right now
Since nothing resets, the one live signal worth tracking is which price you are paying this minute. Off-peak rates are exactly half of peak rates. The status below is computed from the official windows — weekdays 01:00–04:00 and 06:00–10:00 UTC are peak; Chinese public holidays are off-peak in full (holidays aren't marked on this clock).
Current time in UTC
Next switch in --:--:-- — —
Batch jobs, evals, and bulk summarisation cost half as much if you schedule them outside the two weekday peak windows. Interactive traffic usually can't wait — but anything queued can.
02 — THE REAL LIMITSConcurrency, not a calendar
DeepSeek's rate limiting answers one question: how many of your requests are running right now? The ceiling is set per account and per model — every API key under the account shares the same pool, so a second key buys you no extra headroom.
| Model | Concurrency limit | Input / 1M tokens (cache miss) | Output / 1M tokens | Input / 1M tokens (cache hit) |
|---|---|---|---|---|
| DeepSeek-V4.1-Flash deepseek-flash vision |
2,500 concurrent requests | $0.30 peak · $0.15 off-peak | $1.20 peak · $0.60 off-peak | $0.006 peak · $0.003 off-peak |
| DeepSeek-V4-Pro-0813 deepseek-v4-pro |
500 concurrent requests | $1.32 peak · $0.66 off-peak | $3.96 peak · $1.98 off-peak | $0.044 peak · $0.022 off-peak |
Both models offer a 1M-token context window and up to 384K tokens of output, with thinking and non-thinking modes, JSON output, and tool calls. Only Flash accepts image input.
- No RPM / TPM / daily caps are published. There is no requests-per-minute number, no tokens-per-minute number, and no daily token allowance in DeepSeek's official pricing or rate-limit documentation — concurrency is the whole mechanism.
- What a 429 means here: under load, DeepSeek dynamically throttles — requests may queue on keep-alive connections, or return HTTP 429 with a Retry-After header. Capacity “resets” the instant your in-flight count drops; there is no penalty box and no cooldown timer.
- No subscription, no tiers. The API is pure pay-per-token from a prepaid balance, and there is no paid plan that raises the concurrency ceiling.
- Balance is the other hard stop. When your balance reaches zero, calls fail with HTTP 402 (Insufficient Balance). There is no automatic overage charge — you top up, or you stop.
- Deduction order: granted (promotional) balance is always spent first, then topped-up balance.
03 — PRICING WINDOWSPeak & off-peak: the full weekly pattern
Peak pricing applies for 35 of the week's 168 hours. Every other hour — all of Saturday and Sunday, and Chinese public holidays in full — is billed at the off-peak rate, exactly half.
| Window | UTC | Beijing (UTC+8) | Rate |
|---|---|---|---|
| Peak — morning block 1 Mon–Fri |
01:00 – 04:00 UTC | 09:00 – 12:00 | Full (peak) price |
| Peak — morning block 2 Mon–Fri |
06:00 – 10:00 UTC | 14:00 – 18:00 | Full (peak) price |
| Off-peak all other hours |
Weekdays outside the blocks above, all weekend hours, and Chinese public holidays in full | Same pattern, shifted +8 hours | Half price on every rate — input, output, cache hits |
The weekly rhythm: the last peak block of the week ends Friday 10:00 UTC, and the next one opens Monday 01:00 UTC. That 63-hour stretch — plus every weekday gap between 04:00–06:00 UTC and after 10:00 UTC — is when heavy workloads cost half. Nothing about your limits changes at these switches; only the price does.
04 — MODEL CHANGESRetired IDs & what's new in 2026
If old DeepSeek code stopped working this year, it wasn't a quota problem — the model names underneath it were retired.
Retired July 24, 2026, 15:59 UTC: the legacy model IDs deepseek-chat and deepseek-reasoner are fully inaccessible. Any configuration, script, or app still calling them fails outright. Migrate to deepseek-flash (or deepseek-v4-pro). The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted for now and route to V4.1-Flash at Flash prices.
- V4.1-Flash went GA on September 10, 2026 as deepseek-flash; the peak/off-peak dual rate card took effect the same day (04:00 UTC). The previous V4-Flash (0731) was retired.
- V4-Pro stayed. DeepSeek announced it would route V4-Pro traffic to Flash from September 14, 2026, then reversed the decision — deepseek-v4-pro remains available with unchanged billing.
- Search & Fetch APIs are new standalone APIs billed at $0.75 per 1,000 successful requests. They are free until 00:00 Beijing time, November 1, 2026 — after that, per-request billing starts.
- The chat app stays free. DeepSeek Chat (chat.deepseek.com and the mobile apps) has no paid subscription tier. Peak-time throttling applies, but no message cap — and no daily reset — is published.
05 — BALANCE & EXPIRYNothing resets — but does anything expire?
A prepaid balance has no cycle, but expiry is a fair question. Here DeepSeek's documentation is thinner than most: it publishes deduction order, not expiry dates.
- Topped-up balance: DeepSeek publishes no expiry for purchased balance, and past coverage of capacity freezes confirms purchased credits stay with users and are not cancelled. Treat “no known expiry” as the current state — not a contractual promise.
- Granted balance: the pricing page confirms a granted balance exists and is spent before topped-up balance, but publishes no amount, no expiry, and no eligibility rule. The “5M free tokens for 30 days” figure repeated around the web is user-reported only; DeepSeek has never officially stated it.
- No monthly reset of any kind: there is no billing-cycle boundary at which balances, limits, or allowances refresh. Your balance on the 1st of the month is whatever you left on the 31st, minus what you spent.
06 — YOUR NUMBERSWhere to check your own usage & balance
With no reset to track, the console is about two numbers: your balance, and how close to the concurrency ceiling your traffic runs.
DeepSeek console
platform.deepseek.comYour account home for usage history, billing, top-ups, and API key management. This is where any granted balance shows up, and where you refill before a 402 stops production traffic. Top-up accepts Alipay and WeChat Pay; card options vary by account.
Open the console ↗Official pricing & limits
api-docs.deepseek.com/quick_start/pricingThe public source of truth for per-token rates (peak and off-peak), the concurrency limits, the legacy-ID retirement notice, and the peak-hours definition quoted on this page.
Open the pricing page ↗API base (OpenAI-compatible)
api.deepseek.comThe API endpoint itself. Keys are created in the console; all keys on one account draw on the same balance and share the same concurrency pool — splitting keys across services doesn't multiply your limits.
07 — FAQDeepSeek limit questions, answered straight
Does DeepSeek have a quota reset?
No. DeepSeek publishes no daily, weekly, or monthly usage quota, so there is nothing that resets on a schedule. API access is limited by concurrency — how many requests are in flight at once — and by your prepaid balance. The only recurring time-based event on the platform is the switch between peak and off-peak pricing.
What actually limits my DeepSeek API usage?
Two things. First, the account-level concurrency limit: 2,500 simultaneous requests for deepseek-flash and 500 for deepseek-v4-pro. Second, your balance. DeepSeek publishes no requests-per-minute, tokens-per-minute, or daily token caps, and all API keys under one account share the same concurrency pool — extra keys don't raise it.
What happens when I hit the concurrency limit?
Under load, DeepSeek dynamically throttles: requests may queue on keep-alive connections, or fail with HTTP 429 and a Retry-After header. There is no cooldown timer to wait out — capacity returns the moment your in-flight requests finish and the count drops below the ceiling. Client-side, the fix is a concurrency cap and retry-with-backoff, not waiting for a reset that doesn't exist.
When is DeepSeek cheapest to use?
Off-peak — rates are exactly half. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours, including all weekend hours and Chinese public holidays in full, are off-peak. The live status at the top of this page shows which you're in right now.
Does my DeepSeek balance expire or reset monthly?
There is no monthly billing cycle and no reset: billing is a prepaid balance deducted per token, per request. DeepSeek publishes no expiry for topped-up balance. Granted (promotional) balance is spent first; its amount and expiry are not officially published, so check your console. When the balance reaches zero, calls fail with HTTP 402 Insufficient Balance — there is no automatic overage charge.
Is there a free tier for the DeepSeek API?
No documented one. The pricing page mentions a granted balance that is spent before topped-up balance, but publishes no amount, expiry, or eligibility rule. A widely repeated claim of “5M free tokens for 30 days” is user-reported only, with no official DeepSeek source — treat any grant in your console as a bonus, not a promise. Separately, the DeepSeek Chat app is free, with peak-time throttling and no published message cap.
Why did deepseek-chat and deepseek-reasoner stop working?
DeepSeek retired both legacy model IDs on July 24, 2026 at 15:59 UTC; they are fully inaccessible, and any configuration still using them fails. Migrate to the current IDs: deepseek-flash (DeepSeek-V4.1-Flash) or deepseek-v4-pro (DeepSeek-V4-Pro-0813). It was a retirement, not an outage or a quota event.
Does the DeepSeek Chat app have a daily message limit?
No published one. The chat app at chat.deepseek.com and the mobile apps are free, with no paid subscription tier. Peak-time throttling applies, but DeepSeek publishes no message cap and no daily reset for chat — claims of a specific daily number (e.g. “50 messages/day”) come from stale third-party trackers, not from DeepSeek.
Where can I check my DeepSeek usage and balance?
In the console at platform.deepseek.com — usage, billing, top-up, and API keys live there. Current rates and the concurrency limits are published on the official pricing page at api-docs.deepseek.com/quick_start/pricing.
What are DeepSeek's Search and Fetch APIs?
Newer standalone APIs billed per request at $0.75 per 1,000 successful requests. They are free until 00:00 Beijing time on November 1, 2026, when per-request billing starts — worth flagging if you're building on them now.