OFFICIAL PRICES + YOUR ASSUMPTIONS

Cost & Routing

Estimate workload economics without presenting modeled output as observed production telemetry.

Price source verified 2026-07-27
Modeled monthly cost$129.64100,000 requests
Cost per accepted workflow$0.0015385,000 accepted (assumption)
Cache savings$65.48vs identical workload with 0% cached input
Route comparison-47.1%vs GPT-5.6 Luna

Modeled monthly cost breakdown

Exact price components for the selected model route. Fallback traffic is included in each token component.

DeepSeek V4 Pro$0.435/M input · $0.003625/M cached · $0.87/M output
Official source

Sensitivity

Modeled USD per 1,000 accepted workflows. Click a cell to inspect a scenario.

Acceptance ↓ / Cache →
0%20%40%50%60%70%80%
95%
90%
85%
80%
75%
Selected: 50% cached input and 85% acceptance. These are assumptions, not TokenAir observations.

Candidate comparison

Same workload and acceptance assumption; only official model prices change.

ScenarioModelInput / 1MOutput / 1MModeled monthly costAccepted workflowsCost / acceptedEvidence
Selected routeDeepSeek V4 ProDeepSeek$0.44$0.87$129.6485,000$0.00153Official docs
Candidate 1DeepSeek V4 FlashDeepSeek$0.14$0.28$91.7785,000$0.00108Official docs
Candidate 2Gemini 3.5 Flash-LiteGoogle Gemini$0.30$2.50$152.1485,000$0.00179Official docs
Candidate 3Codestral 25.08Mistral AI$0.30$0.90$120.6085,000$0.00142Official docs
How to read thisThe calculator is real arithmetic over official list prices; it is not a production forecast until you replace every default with measured workload data.

Start by measuring accepted workflows, retries and cacheable input in your own logs. Then compare the exported scenario with your invoice.

Review methodology