THE TICKER · DEEPSEEK · CHECKED DAILY · LAST CHANGE 2026-09-10
DEEPSEEK API PRICING
The price disrupter — cheapest input AND output, twice over.
DeepSeek undercuts the priciest model on the board by roughly 23x on input and keeps the only 2x output multiple in the business — on DeepSeek the output is cheap enough that verbose models stop being a bill problem. That flat-rate era ends 2026-08-16, when DeepSeek splits pricing into peak hours (01:00–04:00 and 06:00–10:00 UTC — Beijing office hours) and off-peak, both above today's rate. The catch is not price, it's context: DeepSeek's cost advantage shrinks as your prompts grow, because long-context processing has its own floor.
THE DEEPSEEK LINEUP
USD PER MILLION TOKENS · 2026-10-13| Model | Input / MTok | Output / MTok | Multiple | Cached read | Cache write |
|---|---|---|---|---|---|
| DeepSeek V4 Pro Off-peak rate. Peak (01:00–04:00 and 06:00–10:00 UTC, Monday–Friday; all other hours are off-peak) doubles it: $1.32 in / $3.96 out. DeepSeek cancelled the planned September 2026 retirement of V4 Pro and keeps serving it "with the billing method remaining unchanged". | $0.660 | $1.98 | 3.0x | $0.022 | $0.000 |
| DeepSeek V4.1 Flash DeepSeek retired V4 Flash and routes this same model ID to V4.1 Flash instead — this is what a real call actually bills today, not the old $0.22/$0.66. Off-peak rate, in force 17 of every 24 hours. Peak (01:00–04:00 and 06:00–10:00 UTC) doubles it: $0.30 in / $1.20 out. Cache hit: $0.003/MTok off-peak. | $0.150 | $0.600 | 4.0x | $0.003 | $0.000 |
Source: api-docs.deepseek.com/quick_start/pricing · last confirmed against the daily source on 2026-09-13 · effective public price, promos included.
ONE JOB ACROSS THE DEEPSEEK LINEUP
Estimates, not invoices.
DEEPSEEK CHANGELOG
DeepSeek cancels the V4 Pro retirement it had set for 2026-09-14: the model stays on the API "with the billing method remaining unchanged". The 2026-09-10 note below no longer applies.
DeepSeek quietly retires V4 Flash — the same model ID now serves V4.1 Flash instead, at $0.15 in / $0.60 out per MTok off-peak (-32% in, -9% out). Caught via a LiteLLM pricing bug we filed three weeks ago.
DeepSeek V4 Pro retires 2026-09-14 (12:00 Beijing time) — every call reroutes to V4.1 Flash and bills at its rate from then on. Nothing changes here until that date.
DeepSeek V4 Pro climbs to $0.66 in / $1.98 out per MTok off-peak (+52% in, +128% out), and doubles again at peak. Cache reads went from $0.003625 to $0.022 per MTok, a 6x that nobody put in a headline.
DeepSeek V4 Flash climbs to $0.22 in / $0.66 out per MTok off-peak (+57% in, +136% out), and doubles again at peak.
DeepSeek V4 Pro goes hourly on 2026-08-16: $0.66 in / $1.98 out off-peak, $1.32 / $3.96 at peak — up from a flat $0.435 / $0.87.
DeepSeek V4 Flash goes hourly on 2026-08-16: $0.22 in / $0.66 out off-peak, $0.44 / $1.32 at peak — up from a flat $0.14 / $0.28.
DeepSeek V4 Pro joins the board at $0.435 in / $0.87 out per MTok.