DeepSeek API Pricing 2026: V3 & R1 Costs (Cheapest LLM)
Per-million-token pricing for the DeepSeek API lineup — V3 and R1. DeepSeek is the cheapest capable LLM APIin this dataset, with V3 at roughly one-tenth of GPT-4o's input cost. Cross-compare on the main calculator.
DeepSeek V3 vs GPT-4o — unique cost comparison
On the same per-million-token basis, DeepSeek-V3 is 89% cheaper on input and 89% cheaper on output than GPT-4o. For a workload spending $1,000/month on GPT-4o, switching to V3 drops the bill to roughly $110/month at list price.
Cheapest DeepSeek model for coding
DeepSeek-V3 — $0.27 input / $1.10 output per million tokens. Among the cheapest code-capable APIs from any provider.
DeepSeek pricing per million tokens
USD per 1,000,000 tokens. DeepSeek does not publish separate cached/batch list rates in this dataset; cached promotions change frequently — see the FAQ below.
| Model | Input $/M | Output $/M | Context |
|---|---|---|---|
| DeepSeek-V3 | $0.27 | $1.10 | 128K |
| DeepSeek-R1 | $0.55 | $2.19 | 128K |
DeepSeek-V3
$0.27 in / $1.10 out · 128K context
DeepSeek-R1
$0.55 in / $2.19 out · 128K context
Cheapest capable LLM API — positioning
- Frontier quality at budget pricing: V3 benchmarks competitively with models 5–10× its price.
- Reasoning without the premium: R1 delivers chain-of-thought reasoning at a fraction of OpenAI o3 / Claude Opus 4 cost.
- Trade-offs to plan for: single-provider dependency, occasional rate limits during demand spikes, and data-residency review for sensitive workloads.
Compare DeepSeek against other providers
Frequently asked questions
How much does DeepSeek V3 cost?
DeepSeek-V3 lists at $0.27 per million input tokens and $1.10 per million output tokens. That is roughly 89% cheaper than GPT-4o on input ($2.50) and 89% cheaper on output ($10.00), making it the cheapest frontier-class API in this dataset.
Is DeepSeek R1 good for reasoning?
Yes. DeepSeek-R1 is the reasoning-focused model in the lineup, priced higher than V3 at $0.55/M input and $2.19/M output but still roughly 78% cheaper than OpenAI o3 ($10/$40). For multi-step math, logic and planning workloads where you want frontier-class reasoning without frontier-class pricing, R1 is one of the strongest value picks on the market.
Why is DeepSeek so cheap?
DeepSeek publishes aggressively low list prices as a share-capture strategy, supported by efficient mixture-of-experts architectures and lower-margin positioning than Western labs. The trade-off is provider risk (single-region availability, occasional rate limits during demand spikes) and data-residency considerations. For non-sensitive workloads, the cost saving is substantial.
DeepSeek V3 vs R1 — which should I pick?
Pick V3 ($0.27/$1.10) for general chat, code generation and most production tasks. Pick R1 ($0.55/$2.19) when you specifically need extended chain-of-thought reasoning — competition math, formal logic, complex planning. R1 is about 2× the price of V3 but still dramatically cheaper than Western reasoning models.
Does DeepSeek offer prompt caching or batch discounts?
DeepSeek has historically offered automatic prompt caching on repeated prefixes at steep discounts (reported as low as $0.014/M for cached V3 input during off-peak windows). The rates shown in the table below are standard on-demand list prices; check the official DeepSeek pricing page for current cached and batch promotions, which change frequently.