DeepSeek Models API Cost Calculator & Comparison

Every DeepSeek model, side by side — current API rates, context window, benchmarks, and a live calculator that ranks them at your exact workload. 10 active models, 10 with public pricing. Prices refreshed daily.

Models tracked

10

Active

10

With public pricing

10

Cheapest input

$0.15/1M

Calculate your DeepSeek API cost at your workload.

Set your workload — every priced model ranks in real time.

Adjust the workload

Every model below updates in real time.

1,00010,00050,000250,0001M10M

Ranked by your monthly bill

No models with public pricing available to compare right now.

Pricing at a glance

Blended $/1M tokens across the lineup.

Blended price uses a 3-to-1 input/output ratio. Green bar = cheapest.

Quick picks

Best DeepSeek model for your use case.

As of April 2026, DeepSeek offers 10 active models via API, ranging from $0.15/1M to $0.70/1M input tokens. The most context-rich model handles up to 164K tokens. Models support deep reasoning, tool use. All prices are USD per 1 million tokens.

Lowest cost

DeepSeek V3.1

$0.15/1M input

Largest context

DeepSeek V3.2 Speciale

164K tokens

Deep reasoning

DeepSeek V3.2

Reasoning mode

Tool use / agents

DeepSeek V3.2 Exp

Function calling

Quality vs price

DeepSeek benchmarks at a glance.

Each point is one model — X is blended $/1M tokens, Y is the average of available quality benchmarks. Larger bubbles mean larger context windows.

Per-model benchmark scores

ModelAvgScores
R1 052875.2
MMLU90.8GPQA Diamond81AIME 202491.4AIME 202587.5LiveCodeBench73.3AA Intelligence Index27
DeepSeek V3 032456.0
MMLU Pro81.2GPQA Diamond68.4AIME 202459.4LiveCodeBench49.2AA Intelligence Index22
R1 Distill Qwen 32B55.2
GPQA Diamond62.1MATH94.3AIME 202472.6LiveCodeBench57.2IFEval41.9BBH17.1MMLU Pro41.0
R1 Distill Llama 70B55.1
GPQA Diamond65.2MATH94.5AIME 202486.7LiveCodeBench57.5AA Intelligence Index16IFEval43.4BBH35.8MMLU Pro41.6
DeepSeek V354.9
MMLU88.5HumanEval83AA Intelligence Index16GPQA Diamond67.6SciPredict19.2
DeepSeek V3.239.2
AA Intelligence Index32FrontierMath Tier-42.1GPQA Diamond83.4
DeepSeek V3.2 Exp32.0
AA Intelligence Index32
DeepSeek V3.1 Terminus28.0
AA Intelligence Index28
DeepSeek V3.128.0
AA Intelligence Index28

Open weights

Open Models from DeepSeek

DeepSeek ships 10 open-source or open-weights models you can self-host or fine-tune. Each links to its Hugging Face card.

Every model

Every DeepSeek model — pricing, context & capabilities.

ModelContextInput /1MOutput /1M
DeepSeek V3.2 Speciale164K$0.4$1.20
DeepSeek V3.2131K$0.252$0.378
DeepSeek V3.2 Exp164K$0.27$0.41
DeepSeek V3.1 Terminus164K$0.21$0.79
DeepSeek V3.133K$0.15$0.75
R1 0528164K$0.5$2.15
DeepSeek V3 0324164K$0.2$0.77
R1 Distill Llama 70B131K$0.7$0.8
R1 Distill Qwen 32B33K$0.29$0.29
DeepSeek V3164K$0.32$0.89

FAQ

अक्सर पूछे जाने वाले प्रश्न

Pricing patterns, best-known use cases, and how this provider stacks up.

Get instant answers from our AI agent

DeepSeek API pricing ranges from $0.15 to $0.70 per 1M input tokens. Output tokens cost more than input on every model. Prices are per 1 million tokens (1M ≈ 750,000 words). Use the calculator above to estimate your monthly spend at your actual workload.
DeepSeek V3.1 is the lowest-priced DeepSeek model with public pricing at $0.15/1M input tokens. It suits high-volume tasks where cost matters most — classification, extraction, summarization, and similar workloads that don't need frontier reasoning.
R1 Distill Llama 70B is DeepSeek's highest-tier model at $0.70/1M input. It delivers the most sophisticated reasoning, instruction-following, and nuance. For workloads that don't require frontier performance, a mid-tier model typically cuts inference costs substantially.
DeepSeek V3.2 Speciale, DeepSeek V3.2, DeepSeek V3.2 Exp and 5 more support deep reasoning mode, which improves performance on multi-step coding, debugging, and code review. For simpler autocomplete or snippet generation, a faster, cheaper model often delivers acceptable quality at a fraction of the cost.
DeepSeek V3.2, DeepSeek V3.2 Exp, DeepSeek V3.1 Terminus and 5 more support function calling (tool use), required for agentic workflows. Agents need a model that reliably follows structured output schemas — test with your specific tool definitions before committing to production volumes.
Yes — DeepSeek supports prompt caching (discounts for repeated context) and batch processing (accept a delay, cut costs ~50%). These rates appear in the table above under "Cached /1M" and "Batch /1M." Caching pays off quickly if your prompts share a long system prompt or document prefix across many calls.
DeepSeek has historically adjusted prices when launching new model generations, often cutting rates to stay competitive. Buzzi.ai snapshots pricing daily — you can subscribe to price-drop alerts on any DeepSeek model using the "Alert me" button on its detail page.
Use the main comparison wizard to run the same calculator across DeepSeek, Anthropic, Google, Meta, Mistral, and 20+ other providers. Set your exact workload and get a ranked cost chart in under a minute.
DeepSeek V3.2 Speciale, DeepSeek V3.2, DeepSeek V3.2 Exp, DeepSeek V3.1 Terminus and 4 more offer an extended thinking or reasoning mode. The model spends extra compute "thinking" before answering — slower and more expensive, but meaningfully better on complex, multi-step problems. Standard mode is faster and cheaper for routine tasks.

Look wider

Compare DeepSeek against other providers.

Open the full wizard — pick a use case, set your usage, and cross-compare against OpenAI, Anthropic, Google, and 20+ more.

Open the comparison wizard