DeepSeek API pricing

DeepSeek runs a small, entirely Budget-tier lineup — the V4 Pro and V4 Flash models — and consistently prices among the lowest of any provider we track while still delivering competitive capability on independent benchmarks. See the live table above for current rates.

Why DeepSeek is priced so aggressively

DeepSeek has built its reputation on training-efficiency research, and that efficiency shows up directly in API pricing rather than only in headline benchmark scores. For teams running high-volume, cost-sensitive workloads — classification, summarization, bulk content processing — DeepSeek is frequently the outright cheapest option that still clears a reasonable capability bar, which is why it shows up near the top of the cheapest models ranking across most workload mixes.

Pro vs. Flash

V4 Pro is the more capable of the two and the better default for general-purpose work. V4 Flash trades some capability for an even lower price and higher throughput — suited to very high-volume, simpler tasks where per-request cost matters more than ceiling capability. The price gap between them is smaller than the Budget-to-Premium gaps at larger providers, so testing both directly on your workload (rather than assuming Pro is always worth the difference) is worthwhile here specifically.

What to verify before committing

Because DeepSeek's value proposition rests heavily on price, it's worth confirming the fit deliberately rather than defaulting to it purely on cost:

1. Check benchmark standing for your task type on LiveBench and SWE-bench Verified if coding is a meaningful part of your workload — DeepSeek's coding capability has been competitive, but confirm current standing rather than relying on reputation.
2. Confirm data-handling terms independently for your use case, since providers vary in data residency, retention, and training-use policies — this matters more for regulated or sensitive data than for general prototyping.
3. Test latency and rate limits for your specific volume, since the cheapest per-token price isn't the whole cost picture if throughput constraints force architectural workarounds.

Rate limits

DeepSeek takes a notably different approach from most other providers: there's no published fixed requests-per-minute or tokens-per-minute tier system. Capacity is allocated dynamically based on real-time server load — when demand is light you may see high effective throughput with generous concurrency, and during peak demand periods requests can be throttled more than a hard-tiered provider's guaranteed floor would allow. This trades predictability for potentially higher throughput off-peak — worth knowing if your workload needs a guaranteed floor rather than best-effort capacity.

See how this compares to the other 8 providers we track on the rate limits page.

When DeepSeek vs. a competitor

For pure cost-sensitive routing, DeepSeek is a strong first check alongside Qwen and Moonshot — the three lowest-cost provider lines we track. If your workload needs capability beyond what the Budget tier delivers, compare against Standard-tier options from OpenAI, Google, or Anthropic rather than pushing a Budget-tier model past its comfortable range — see Which Frontier Model Is Actually Cheapest for Your Workload? for how to make that call systematically rather than by price alone.

See current DeepSeek pricing live above, or find where it ranks against every tracked model on cheapest models.