DeepSeek Pricing 2026: Free App, API Rates, and Flash vs Pro

DeepSeek pricing has no subscription tiers. No paid plans exist - not for individuals, not for teams. Both V4-Flash and V4-Pro, including the 1M-token context window and full reasoning capability, are free in the consumer app. API access is the only paid path, billed per token against a prepaid balance with no monthly minimum. Prices verified against DeepSeek's official API docs on July 29, 2026.

Free App: What You Get Without Paying

chat.deepseek.com is fully free. No premium tier, no credit system, and no documented usage cap. Two modes give access to both models: Instant Mode runs V4-Flash (284B total / 13B active parameters), and Expert Mode runs V4-Pro (1.6T total / 49B active parameters). A free flagship reasoning model with a 1M-token context window is rare - most providers reserve their best model for paid subscribers, offering a limited version on the free tier. For a full comparison of free tiers across major AI tools, see our free AI tools guide.

DeepSeek runs on iOS, Android, and the web, with file uploads, web search, and code execution available across all platforms. No hard daily cap appears in the published documentation. Heavy users report throttling at peak periods, but DeepSeek has not confirmed a specific number anywhere in its terms or FAQ.

API Pricing: DeepSeek-V4-Flash

V4-Flash is DeepSeek's fast, cost-efficient API model. Input costs $0.14 per 1M tokens on a cache miss and $0.28 per 1M output tokens. Context caching cuts the input price to $0.0028 per 1M - a 98% reduction from the standard rate, and one of the deepest caching discounts available from any frontier provider. Concurrency tops out at 2,500 simultaneous requests, making Flash the practical choice for batch jobs, agent pipelines, and high-throughput production workloads.

V4-Flash has 284B total parameters, 13B active. A mixture-of-experts architecture keeps inference fast without sacrificing much quality relative to Pro. On simple agent tasks, Flash performance sits close to V4-Pro. For most production use cases, Flash is the right starting point before considering whether V4-Pro's accuracy gains justify a 3x cost increase.

API Pricing: DeepSeek-V4-Pro

V4-Pro is DeepSeek's flagship API model. Input runs $0.435 per 1M tokens on a cache miss, dropping to $0.003625 per 1M on a cache hit. Output costs $0.87 per 1M tokens. Parameters total 1.6T, with 49B active under the same mixture-of-experts architecture - the larger active parameter count is where Pro earns its 3x price premium over Flash.

Per DeepSeek's April 2026 benchmarks, V4-Pro leads all open models on math, STEM, and coding, trailing only Gemini-3.1-Pro on world knowledge. DeepSeek supports both OpenAI-compatible and Anthropic-compatible API endpoints. Concurrency caps at 500 requests - lower than Flash. For a side-by-side comparison of V4-Pro rates against other frontier APIs, see our AI API pricing guide.

Context Caching

Both models include automatic context caching. Cache hits cut Flash input costs by 98%, from $0.14 to $0.0028 per 1M tokens. Pro drops further - from $0.435 to $0.003625 per 1M, a 99% reduction. Caching works best when a shared system prompt, repeated document context, or long multi-turn conversation history stays constant between requests, which covers most RAG and agentic use cases.

Priming the cache requires no separate API call. DeepSeek detects repeated prefixes and applies the reduced rate automatically. For batch jobs, document analysis pipelines, or chat applications with long system prompts, cache hits will be frequent and input costs will fall well below the headline cache-miss price.

The Catches

DeepSeek is a Chinese company based in Hangzhou, and data processed through chat.deepseek.com and the API falls under Chinese data jurisdiction. Model training uses conversation data by default - that clause is in the privacy policy. Enterprises with data residency requirements or privacy-sensitive workloads should review the full terms before committing either the app or the API to production.

API billing runs on a top-up balance. No monthly minimum, no seat fee, no auto-renewal. Volume discounts don't exist - every developer pays the same per-token rate whether running a thousand tokens per month or a billion. For teams at scale, the absence of committed-spend pricing is worth factoring into cost projections early.

Both models default to thinking mode, and thinking tokens bill at the full output rate. For classification, short extraction, or low-complexity generation tasks, chain-of-thought adds no value - but the cost is the same as a complex math query. Switching to non-thinking mode on those calls cuts output costs without changing the model or the endpoint.

V4-Pro's concurrency cap is 500 simultaneous requests. Traffic spikes can breach it. Flash's 2,500 cap handles more burst traffic, but neither model offers a higher rate limit without a direct conversation with DeepSeek's team.

Flash, Pro, or Free App: Which to Use

Free app (chat.deepseek.com) covers research, writing, coding, and document work at no cost, with V4-Pro access included alongside V4-Flash. No subscription required. No financial reason exists to use the API unless programmatic access or production integration is the goal.

V4-Flash API ($0.14/$0.28 per 1M tokens) suits high-throughput applications and cost-sensitive production workloads. Flash handles 2,500 concurrent requests. Context caching keeps repeated-context input costs at $0.0028 per 1M - cheap enough to run at scale even for long-prompt applications where input tokens dominate the bill.

V4-Pro API ($0.435/$0.87 per 1M tokens) fits complex reasoning, frontier-accuracy coding tasks, and multi-step agentic workflows. Cost matters less here than quality. Even at Pro rates, DeepSeek undercuts Claude Fable 5 and GPT-5 by 5-10x for comparable capability - and DeepSeek's recent $7.4 billion raise at a $50 billion valuation suggests this pricing reflects a deliberate market position, not a temporary loss leader.

Context caching applies to both models. Input cost on a cache hit is $0.0028 (Flash) or $0.003625 (Pro) per 1M tokens - a difference that compounds fast in any high-frequency or long-context application. For the full picture of what every major AI provider charges, see our AI pricing comparison.