Deepseek Token API pricing. DeepSeek-V4-Flash, DeepSeek-V4-Pro and 1 more models. OpenAI-compatible API, Anthropic-compatible API, Claude Code and 3 more tools integrations. Includes cache tiers and rate limits. API key setup, usage quotas, and official entry—is it worth it?

DeepSeek API Latest Updates

DeepSeek API Price Increase: How Much It's Going Up, V4-Flash vs V4-Pro Current Pricing and What to Do

Published 2026-08-10Updated 2026-08-10

PricingDeepSeekAPIPricing UpdatePrice IncreaseV4-FlashV4-Pro

Summary

DeepSeek API prices are going up — the official pricing page has confirmed a significant increase is coming. Current rates: V4-Flash cache-miss input $0.14/MTok, output $0.28/MTok; V4-Pro input $0.435/MTok, output $0.87/MTok; cache-hit as low as $0.0028-0.003625/MTok. This guide compares current pricing, provides caching optimization strategies, and covers third-party alternatives like Volcengine and CtCloud so you can plan your budget before the hike hits.

DeepSeek is raising API prices, and it's not a small bump. If you're on their API, you need to do the math now.

The official pricing page in August 2026 dropped a notice: overall API pricing is going up, and the increase is "significant" (source). No exact numbers or dates yet — just a heads-up to plan your usage.

Two models in play right now: deepseek-v4-flash and deepseek-v4-pro. Both support 1M context, up to 384K output, OpenAI and Anthropic format compatibility.

Current Pricing: What You're Paying Now

Get clear on the current rates before they change.

DeepSeek-V4-Flash

Item	Price (USD)	Price (CNY)
Cache-hit input	$0.0028 / MTok	¥0.02 / MTok
Cache-miss input	$0.14 / MTok	¥1 / MTok
Output	$0.28 / MTok	¥2 / MTok

Concurrency 2500. Thinking/non-thinking dual mode, JSON Output, Tool Calls, FIM completion. Responses API already supported.

DeepSeek-V4-Pro

Item	Price (USD)	Price (CNY)
Cache-hit input	$0.003625 / MTok	¥0.025 / MTok
Cache-miss input	$0.435 / MTok	¥3 / MTok
Output	$0.87 / MTok	¥6 / MTok

Concurrency 500. Same feature set as Flash, but Responses API not yet live (expected early August).

Caching Matters Way More Than Model Choice

Seriously — good caching crushes bad model selection every time. V4-series caching drops input costs by 98% or more:

	Flash Cache-Hit	Flash Cache-Miss	Pro Cache-Hit	Pro Cache-Miss
Input price	$0.0028 / MTok	$0.14 / MTok	$0.003625 / MTok	$0.435 / MTok
Savings vs miss	98%	—	99.2%	—

Where caching pays off hardest: repeated system prompts, multi-turn coding sessions, RAG knowledge-base lookups. Cache hit rate is your real cost lever.

How Billing Works

Cost = token usage × model unit price, auto-deducted from your balance. Gift balance gets used first, then purchased. When balance hits zero, the API stops. Top up what you need — don't hoard.

What to Do Before the Hike

Third-party Coding Plans

DeepSeek has no official subscription plan, but Volcengine, CtCloud, and Baidu Qianfan all offer fixed-price plans with DeepSeek model access. These run on plan quotas — separate from official billing, immune to the coming price hike.

Migrating Is Practically Free

DeepSeek API supports both OpenAI and Anthropic formats. In Claude Code, Cursor, Cline, or Codex CLI, you're just swapping a Base URL and API key. That's it.

Bottom Line

Keep hitting https://api-docs.deepseek.com/quick_start/pricing — the actual hike numbers will land there.
Calc your monthly token spend now so you know the financial impact before it hits.
Get caching dialed in today. Don't wake up post-hike with a sub-50% cache hit rate.
Route properly: bulk and daily work → Flash, complex reasoning only → Pro.

Sources and verification

Models & Pricing - DeepSeek API Docs — DeepSeek (2026-08-10)

Last verified：2026-08-10

Deepseek Token API FAQ

What exactly is the DeepSeek API price increase announcement about, and how should I plan my budget given the current V4-Flash and V4-Pro pricing?

DeepSeek's official pricing page (August 2026) explicitly announced a coming overall API price increase — and it's going to be significant. New prices aren't published yet, but you're being told to plan ahead. Current rates: V4-Flash cache-miss input $0.14/MTok, output $0.28/MTok; V4-Pro input $0.435/MTok, output $0.87/MTok. Cache-hit input is dramatically cheaper — Flash at $0.0028/MTok, Pro at $0.003625/MTok. Before the hike hits, get your caching strategy tight — that's where the real savings live.

When does the DeepSeek API price increase take effect, how do cache-hit vs cache-miss costs work, and is my current usage affected?

No date yet — just 'in the near future.' Your current calls are still billed at existing rates. On billing mechanics: gift balance gets used first, then purchased balance. Cache-hit input prices are Flash $0.0028/MTok and Pro $0.003625/MTok — dozens of times cheaper than cache-miss. Keep an eye on https://api-docs.deepseek.com/quick_start/pricing for the actual notice.

How can I use context caching to aggressively cut DeepSeek API input costs for Flash and Pro models after the price hike?

V4-series caching can slash input costs by over 98% — honestly, this matters way more than whether you pick Flash or Pro. Repeated system prompts, long conversation histories, RAG document contexts — these are your highest-ROI caching targets. If you're running Claude Code or your own agent pipeline, track cache hit rate as your primary cost metric. It'll save you more money than obsessing over the pricing page.

After the DeepSeek API price hike, should I stick with Flash or Pro, and are fixed-price third-party coding plans worth switching to?

Don't route everything through Pro after the hike — reserve it for complex reasoning only. Bulk and daily completions go to Flash (concurrency 2500 vs Pro's 500 — that alone makes a big cost difference). Also worth checking: [Volcengine](/en/huoshan-fangzhou-token-plan), [CtCloud](/en/tianyi-token-plan), and [Baidu Qianfan](/en/baidu-token-plan) offer DeepSeek models through fixed-price coding plans with independent pricing — completely insulated from official rate hikes.

What alternatives to the official DeepSeek API channel exist after the price increase, and how do I switch to Volcengine, CtCloud, or Baidu Qianfan coding plans?

If the official hike blows past your budget, third-party coding plans are your fastest exit. [Volcengine](/en/huoshan-fangzhou-token-plan), [CtCloud](/en/tianyi-token-plan), and [Baidu Qianfan](/en/baidu-token-plan) all run DeepSeek models through plan quotas — not official balance. Switching is near-zero effort: DeepSeek API supports both OpenAI and Anthropic formats, so you're just changing a Base URL and API key in Claude Code, Cursor, or Cline.