Models / DeepSeek V4 Pro / Calculator

DeepSeek V4 Pro cost & VRAM calculator

What does DeepSeek V4 Pro cost to run for your workload, and can you run it on your own hardware? Set your workload below - we compute per-provider API cost live and tell you honestly whether local hardware can run it.

LOCAL ISN'T PRACTICAL

This is where owning stops making sense

DeepSeek V4 Pro is a 1600-billion-parameter model. No rig an individual can buy runs it - so unlike a workstation GPU model, there's no break-even to compute. The honest answer for nearly everyone is the per-provider API cost below.

1.6T total, 49B active per token (MoE). Hybrid attention: Compressed Sparse Attention (CSA, 4x KV compression) + Heavily Compressed Attention (HCA, 128x) across 61 layers, manifold-constrained Hyper-Connections (mHC). FP4 experts + FP8 rest, 32T+ pretraining tokens, Muon optimizer. Context: 1M native; Think Max mode recommends >=384K. Self-hosting: ~800GB+ for Q4 weights - needs a multi-H200/H100 cluster or DGX-class system. Not workstation-fittable, so cloud-only for nearly everyone despite the open weights. Top open-weight agentic coding score (SWE-bench Verified 80.6). Open weights under MIT. API phase-out (announced 2026-09-09): from 2026-09-14 04:00 UTC, deepseek-v4-pro requests route to DeepSeek V4.1 Flash at Flash rates (previously $1.32 peak in / $3.96 peak out per 1M) until V4.1-Pro launches.

Your workload

Casual Power user Team of 5 Codebase task

DeepSeek V4 Pro runs an always-on thinking mode. Reasoning (thinking) tokens are billed at the output rate ($15.00/M), so count them here to see the thinking portion of your bill.

API cost for your workload

Provider Rate ($/1M) Monthly cost
Nous Portal may be stale $0.53 in · $1.58 out $0.01 cheapest
Relace $0.90 in · $1.50 out · $0.25 cache $0.02
StreamLake $0.96 in · $1.91 out · $0.08 cache $0.02
GMICloud $0.96 in · $1.91 out · $0.08 cache $0.02
DigitalOcean $1.04 in · $2.09 out · $0.21 cache $0.03
Reka $1.05 in · $2.10 out · $0.21 cache $0.03
Cloudflare $1.15 in · $2.55 out · $0.20 cache $0.03
DeepInfra $1.30 in · $2.60 out · $0.10 cache $0.03
SiliconFlow $1.50 in · $3.14 out · $0.14 cache $0.04
Alibaba $1.56 in · $3.12 out · $0.13 cache $0.04
Novita $1.60 in · $3.20 out · $0.14 cache $0.04
Venice $1.65 in · $3.30 out · $0.33 cache $0.04
AtlasCloud $1.68 in · $3.38 out · $0.13 cache $0.04
Baidu $1.69 in · $3.38 out · $0.14 cache $0.04
Azure $1.91 in · $3.83 out · $0.16 cache $0.05
Crusoe avoid $1.74 in · $3.48 out · $0.15 cache $0.04

Monthly cost is an estimate from list prices and your workload - verify against the provider before committing. Cached fraction applies the cache rate to that share of input.

Can you run it locally?

NO - NO INDIVIDUAL RIG RUNS IT

DeepSeek V4 Pro has no published quantization that fits a rig one person can buy, so there's no local-hardware recommendation and no break-even to compute. The honest answer is the per-provider API cost above.

1.6T total, 49B active per token (MoE). Hybrid attention: Compressed Sparse Attention (CSA, 4x KV compression) + Heavily Compressed Attention (HCA, 128x) across 61 layers, manifold-constrained Hyper-Connections (mHC). FP4 experts + FP8 rest, 32T+ pretraining tokens, Muon optimizer. Context: 1M native; Think Max mode recommends >=384K. Self-hosting: ~800GB+ for Q4 weights - needs a multi-H200/H100 cluster or DGX-class system. Not workstation-fittable, so cloud-only for nearly everyone despite the open weights. Top open-weight agentic coding score (SWE-bench Verified 80.6). Open weights under MIT. API phase-out (announced 2026-09-09): from 2026-09-14 04:00 UTC, deepseek-v4-pro requests route to DeepSeek V4.1 Flash at Flash rates (previously $1.32 peak in / $3.96 peak out per 1M) until V4.1-Pro launches.

See the model card for the full architecture notes and any cloud subscription plans.

Full model card API pricing table Generic token calculator