specification: FinOps Framework
specificationVersion: '1.0'
schema: https://www.finops.org/framework/
alignedWith:
  framework: FinOps Foundation Framework
  frameworkUrl: https://www.finops.org/framework/
  dataSpec: FOCUS
  dataSpecVersion: '1.3'
  dataSpecUrl: https://focus.finops.org/focus-specification/v1-3/
provider: DeepSeek
providerId: deepseek
publisherName: DeepSeek
serviceCategory: AI Infrastructure
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-05'
reconciled: true
tags:
  - AI
  - LLM
  - Token-Based Pricing
  - FinOps
  - FOCUS
description: 'FOCUS-aligned FinOps for DeepSeek: pure pay-as-you-go token-based usage billing across
  two production models (V4-Flash, V4-Pro) with separate meters for cache-hit input, cache-miss input,
  and output tokens.'
sources:
  - https://api-docs.deepseek.com/quick_start/pricing
  - https://api-docs.deepseek.com/quick_start/rate_limit
billingModel:
  pricingCategory: Token-Based Usage
  billingFrequency: On-Demand
  billingCurrency: USD
  chargeCategories:
    - Usage
    - Purchase
    - Adjustment
    - Credit
focusColumns:
  ServiceName: DeepSeek API
  ServiceCategory: AI Infrastructure
  ServiceSubcategory: Large Language Models
  ProviderName: DeepSeek
  PublisherName: DeepSeek
  InvoiceIssuerName: DeepSeek
  PricingCategory: Usage-Based
  PricingUnit: token
  BillingCurrency: USD
  ChargeCategory: Usage
meters:
  - name: tokens_input_cache_hit
    description: Input tokens served from the prompt cache at the discounted cache-hit price.
    unit: token
    aggregation: sum
    dimensions:
      - model
      - api_key
      - consumer
  - name: tokens_input_cache_miss
    description: Input tokens computed fresh (cache miss) at the full input price.
    unit: token
    aggregation: sum
    dimensions:
      - model
      - api_key
      - consumer
  - name: tokens_output
    description: Output tokens generated by the model.
    unit: token
    aggregation: sum
    dimensions:
      - model
      - api_key
      - consumer
  - name: api_requests
    description: Count of completed Chat Completions or FIM requests.
    unit: request
    aggregation: count
    dimensions:
      - model
      - endpoint
      - api_key
principles:
  - name: Visibility
    description: Track spend through the DeepSeek User Balance API and per-request token usage
      returned in each response payload. There is no native cost-explorer dashboard, so consumers
      typically aggregate response metadata into their own observability stack.
  - name: Allocation
    description: Issue separate API keys per team, environment, and product surface so token usage
      can be sliced by api_key. DeepSeek does not provide native cost tags.
  - name: Optimization
    description: Maximize cache-hit ratio by stable prompt prefixes (system prompts, few-shot
      examples, RAG context) — cache hits are roughly 50x cheaper than cache misses on V4-Flash and
      ~120x cheaper on V4-Pro. Prefer V4-Flash for non-reasoning workloads and reserve V4-Pro for
      tasks that benefit from thinking mode. Capitalize on the V4-Pro 75% promotional discount
      through 2026-05-31.
  - name: Accountability
    description: Set per-key spend alerts via the Balance API and rotate keys per business unit so
      overage incidents are attributable to a budget owner. Throttle internally when balance
      drops below a configured threshold.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com