DeepSeek vs OpenRouter pricing: who's cheaper? 核验日期 2026-09-17

DeepSeek models are sold by DeepSeek itself, by SiliconFlow, and through aggregators like OpenRouter. Same weights, different bills. Here are the verified numbers — including the traps (off-peak windows, cache pricing, aggregator markups).

DeepSeek-class prices by seller per 1M tokens

Seller	Model	Input $/1M	Cache read $/1M	Output $/1M	Notes
DeepSeek (official)	deepseek-flash (V4.1 Flash)	$0.30	$0.006	$1.20	Peak. Off-peak ≈ half: $0.15/$0.60, cache-hit $0.3¢
SiliconFlow	DeepSeek-V4-Flash	$0.13	$0.028	$0.28	No off-peak needed — cheaper than official off-peak output
OpenRouter	deepseek-v4-flash	$0.07	$0.014	$0.14	Current-gen V4 Flash — cheapest DeepSeek-class listing we verified (via OpenRouter's public models API)
OpenRouter	deepseek-v4.1-flash	$0.15	$0.003	$0.60	≈ official off-peak, 24/7
DeepSeek (official)	deepseek-v4-pro	$1.32	$0.044	$3.96	Peak. Off-peak $0.66/$1.98
SiliconFlow	DeepSeek-V4-Pro-0813	$1.32	$0.044	$3.96	Same as official peak
OpenRouter	deepseek-v3.2 (deepseek-chat)	$0.2574	待核验	$1.0287	Older-gen V3.2, still on the menu

The traps

Off-peak windows: official DeepSeek peak hours are 01:00–04:00 & 06:00–10:00 UTC Mon–Fri; the rest of the week bills at ½ peak. Nightly CI batches can legally ride the half-price window.
Cache pricing isn't comparable across sellers: DeepSeek's cache-hit rate ($0.006/1M peak) is dramatically cheaper than SiliconFlow's ($0.028/1M) — but only if your workload actually hits cache. Agent workloads usually do.
Aggregators trade price for convenience: OpenRouter's one-key-any-model convenience costs little here, but its V3.2 listing is last-gen. For current-gen V4 Flash it's the opposite — OpenRouter's deepseek-v4-flash ($0.07/$0.14) undercuts every other seller we checked, SiliconFlow included.
1M context everywhere: all rows above are 1M-context models — no premium gatekeeping between sellers on context length.

Verdict

Cheapest verified DeepSeek-class route: OpenRouter deepseek-v4-flash — $0.07/$0.14 with cache read at $0.014, verified via OpenRouter's public models API. SiliconFlow ($0.13/$0.28) is the cheapest first-party-style alternative.
Cheapest if you can shift schedules: official DeepSeek off-peak ($0.15/$0.60 output) with cache-hit at $0.003/1M — for nightly batch/CI jobs, effectively free input.
One key for everything: OpenRouter is now both the cheapest and the simplest route for V4 Flash — a rare combination; re-verify before long-term budgeting since aggregator prices drift.
Claude Code users: DeepSeek serves the Anthropic protocol at api.deepseek.com/anthropic — see /configs.

FAQ

Is OpenRouter more expensive than going direct?: It can be cheaper — two listings apply: deepseek-v4-flash ($0.07/$0.14) is the cheapest DeepSeek-class route we verified anywhere, while deepseek-v4.1-flash ($0.15/$0.60) matches official off-peak pricing around the clock. Aggregator prices can drift per underlying provider.
What happened to deepseek-chat / deepseek-reasoner?: DeepSeek's docs now sell deepseek-flash (V4.1 Flash) and deepseek-v4-pro; older aliases still resolve but are billed at the Flash price. There's also an official Anthropic-format endpoint at /anthropic.
Do prices include taxes?: Prices exclude applicable taxes (stated on official pages). Aggregator prices can also drift per underlying provider.
When was this verified?: 2026-09-17 — from api-docs.deepseek.com, siliconflow.com/pricing and OpenRouter's public models API. Re-verify before long-term budgeting.