{"id":2100,"date":"2026-08-03T18:32:03","date_gmt":"2026-08-03T18:32:03","guid":{"rendered":"https:\/\/convly.ai\/deepseek-api-pricing\/"},"modified":"2026-08-03T18:32:03","modified_gmt":"2026-08-03T18:32:03","slug":"deepseek-api-pricing","status":"publish","type":"page","link":"https:\/\/convly.ai\/pt\/deepseek-api-pricing\/","title":{"rendered":"Pre\u00e7os da API DeepSeek (2026): Custo por 1 milh\u00e3o de tokens para cada modelo"},"content":{"rendered":"

DeepSeek API pricing starts at $0.14 per 1M input tokens on DeepSeek V4-Flash \u2014 around
\na sixtieth of what frontier models charge.<\/strong> That gap, not the parameter counts, is why
\nDeepSeek matters. Every model in the line is also released under an open licence, so the API
\nprice is a convenience charge rather than the only way to run them.<\/p>\n

\n

DeepSeek API pricing: every model, per 1M tokens<\/h2>\n \n \n \n \n \n \n \n
Model<\/th>Input $\/1M<\/th>Output $\/1M<\/th>\n Blended $\/1M<\/th>Context<\/th>\n <\/tr><\/thead>\n
DeepSeek V4-Flash<\/a> open weights<\/span><\/td>\n $0.140<\/td>\n $0.280<\/td>\n $0.168<\/strong><\/td>\n 1M<\/td>\n <\/tr>\n
DeepSeek V4-Pro<\/a> open weights<\/span><\/td>\n $0.435<\/td>\n $0.870<\/td>\n $0.522<\/strong><\/td>\n 1M<\/td>\n <\/tr>\n
DeepSeek R1 Distill Llama 70B<\/a> open weights<\/span><\/td>\n $0.800<\/td>\n $0.800<\/td>\n $0.800<\/strong><\/td>\n 128K<\/td>\n <\/tr>\n
DeepSeek R1<\/a> open weights<\/span><\/td>\n $0.500<\/td>\n $2.15<\/td>\n $0.830<\/strong><\/td>\n 128K<\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n

Blended is the effective rate at a 4:1 input-to-output mix, which is what a\n typical chat or retrieval workload actually produces. It is the number to compare across\n vendors \u2014 a headline input price hides how much the output side costs.<\/p>\n\n

What DeepSeek costs per month<\/h2>\n \n \n \n \n \n
Workload<\/th>Tokens \/ month<\/th>\n DeepSeek V4-Flash
cheapest<\/span><\/th>\n 		DeepSeek R1
most capable tier<\/span><\/th>\n <\/tr><\/thead>\n
Side project<\/td>\n 1M in \/ 0.25M out<\/td>\n $0.21<\/strong><\/td>\n $1.04<\/strong><\/td>\n <\/tr>\n
Small team<\/td>\n 20M in \/ 5M out<\/td>\n $4.20<\/strong><\/td>\n $21<\/strong><\/td>\n <\/tr>\n
Production<\/td>\n 200M in \/ 50M out<\/td>\n $42<\/strong><\/td>\n $208<\/strong><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n

Model your own volumes in the\n AI API cost calculator<\/a>.<\/p>\n\n

How DeepSeek compares to other providers<\/h2>\n

Each provider's cheapest priced model, so the comparison is like for like on entry cost.<\/p>\n \n \n \n \n \n \n \n \n \n \n \n \n \n
Provider<\/th>Cheapest model<\/th>Blended $\/1M<\/th><\/tr><\/thead>\n
Mistral AI<\/td>\n Mistral 7B<\/a><\/td>\n $0.0220<\/td>\n <\/tr>\n
Meta<\/td>\n Llama 3.1 8B<\/a><\/td>\n $0.0220<\/td>\n <\/tr>\n
Alibaba<\/td>\n Qwen3 8B<\/a><\/td>\n $0.0600<\/td>\n <\/tr>\n
Google<\/td>\n Gemma 3 4B<\/a><\/td>\n $0.0600<\/td>\n <\/tr>\n
Microsoft<\/td>\n Phi-4<\/a><\/td>\n $0.0840<\/td>\n <\/tr>\n
DeepSeek this page<\/span><\/td>\n DeepSeek V4-Flash<\/a><\/td>\n $0.168<\/td>\n <\/tr>\n
Moonshot AI<\/td>\n Kimi K2.7 Code<\/a><\/td>\n $0.980<\/td>\n <\/tr>\n
Anthropic<\/td>\n Claude Haiku 4.5<\/a><\/td>\n $1.80<\/td>\n <\/tr>\n
Zhipu AI<\/td>\n GLM 5.2<\/a><\/td>\n $2.00<\/td>\n <\/tr>\n
xAI<\/td>\n Grok 4<\/a><\/td>\n $5.40<\/td>\n <\/tr>\n
OpenAI<\/td>\n GPT-5.6 Sol<\/a><\/td>\n $10.00<\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n

Cheapest is not the same as best value \u2014 check capability alongside price on the\n LLM leaderboard<\/a>, or browse every\n model in the AI models database<\/a>.<\/p>\n <\/div>\n\n

Which DeepSeek model should you use?<\/h2>\n

V4-Flash<\/strong> is the volume workhorse: 284B total parameters with roughly 13B
\nactive per token, a 1M-token context, and a blended rate near $0.17 per 1M against a frontier
\nmodel’s $10. For bulk classification, document processing, first-pass summarisation, synthetic
\ndata generation and the retrieval layer of a RAG pipeline, it is close to unmatched on
\ncapability per dollar. A 1M context at that price does not exist elsewhere.<\/p>\n

V4-Pro<\/strong> is the open flagship \u2014 a 1.6-trillion-parameter mixture-of-experts
\nactivating around 49B parameters per token, with a 1M-token context. At roughly $0.52 blended it
\nruns about a twentieth of frontier pricing while holding its own on general reasoning.<\/p>\n

R1<\/strong> is the landmark open reasoning model, a 671B mixture-of-experts with 37B
\nactive, delivering chain-of-thought reasoning at a fraction of proprietary cost. Its 128K
\ncontext is narrower than the V4 line’s 1M.<\/p>\n

R1 Distill Llama 70B<\/strong> exists for one reason: hardware. Full R1 needs roughly
\n400 GB of VRAM at 4-bit; the distill needs about 40 GB, which moves it from “rent a server” to
\n“buy a workstation”. It also bills $0.80 for both input and output, so generation-heavy
\nworkloads cost no more than prompt-heavy ones \u2014 unusually simple to budget.<\/p>\n

The switching cost is lower than you think<\/h2>\n

The detail that makes DeepSeek a genuine option rather than a curiosity is that the V4-Pro
\nAPI speaks both OpenAI and Anthropic request formats. Migrating is usually a base-URL and key
\nchange rather than a rewrite. The integration friction that normally protects incumbent
\nproviders largely disappears, which is what makes a 20x cost difference actionable instead of
\ntheoretical.<\/p>\n

The sensible pattern is not wholesale migration but routing: send the high-volume,
\nlow-stakes majority of traffic to V4-Flash and keep a frontier model for the requests that
\ngenuinely need it. Most applications discover that the split is heavily weighted toward the
\ncheap side.<\/p>\n

Do the open weights change the maths?<\/h2>\n

Every DeepSeek model here is openly downloadable, which raises the obvious question. For
\nmost teams the honest answer is that the licence buys auditability and provider portability
\nrather than a realistic on-premises deployment.<\/p>\n

The hardware requirements are the reason. V4-Pro needs roughly 800 GB of VRAM at 4-bit \u2014 a
\ndata-centre exercise of eight H100 80GBs or more. R1 needs about 400 GB. V4-Flash needs about
\n140 GB, or two H100 80GBs. Only the R1 distill, at 40 GB, fits hardware a small team would
\nactually buy.<\/p>\n

Set against API pricing this low, self-hosting has to clear a high bar: it only pays once
\nyour volume is high enough to keep that hardware continuously busy, and the break-even point
\nmoves every time DeepSeek cuts prices. Work it out for your own token volume in the
\nself-hosting vs API calculator<\/a>
\nbefore committing to hardware. What the open weights do guarantee is that you can never be
\nlocked in, price-gouged, or stranded by a deprecation \u2014 which is worth something even if you
\nnever download them.<\/p>\n

\n

Frequently asked questions<\/h2>\n
\n
How much does the DeepSeek API cost?<\/summary>

DeepSeek API pricing runs from $0.168 per 1M blended tokens on DeepSeek V4-Flash up to $0.830 on DeepSeek R1. Blended assumes a 4:1 input-to-output mix. Input and output are billed separately, and output is always the more expensive side.<\/p><\/details>\n

What is the cheapest DeepSeek model?<\/summary>

DeepSeek V4-Flash, at $0.140 per 1M input tokens and $0.280 per 1M output. A small-team workload of 20M input and 5M output tokens a month costs about $4.20 on it.<\/p><\/details>\n

How much does DeepSeek cost per month?<\/summary>

At 20M input and 5M output tokens a month, DeepSeek V4-Flash costs about $4.20 and DeepSeek R1 about $21. A side project at 1M\/0.25M costs a small fraction of that. Use the AI API cost calculator for your own volumes.<\/p><\/details>\n

Is DeepSeek cheaper than Mistral AI?<\/summary>

On entry-level pricing, DeepSeek starts at $0.168 per 1M blended and Mistral AI starts at $0.0220 on Mistral 7B. Mistral AI is the cheaper entry point, though capability differs \u2014 compare both on the leaderboard before switching.<\/p><\/details>\n

Why does DeepSeek charge more for output tokens than input?<\/summary>

Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost more to serve. This is why prompt-heavy workloads such as retrieval and classification are far cheaper to run than generation-heavy ones, and why the blended rate matters more than the headline input price.<\/p><\/details>\n <\/div>\n

Prices are the published list rates for each model's primary API and are reviewed as\n providers change them. Volume, batch and cached-input discounts are not included. Last reviewed\n August 2026.<\/p>\n<\/div>\n\n