specification: FinOps Framework
specificationVersion: '1.0'
schema: https://www.finops.org/framework/
provider: Together AI
providerId: together-ai
created: '2026-05-08'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-08, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-08'
reconciled: true
tags:
- AI
- LLM
- Inference
- Open Source
- Fine-tuning
- GPU
- FinOps
- Cost Management
- FOCUS
description: >-
  FinOps view of Together AI spend. Together bills usage-based per-token rates
  for serverless inference (chat, embeddings, rerank, vision, audio), per-asset
  rates for image and video, per-1M-character rates for audio, per-token rates
  for fine-tuning training, and hourly rates for dedicated endpoints and GPU
  clusters (with reserved discounts).
notes: >-
  Per-model rates change frequently; verify against the Together pricing page at
  reconciliation. Batch API consistently offers up to 50% off serverless rates.
sources:
- https://www.together.ai/pricing
- https://docs.together.ai/
- https://focus.finops.org/focus-specification/v1-3/
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/
publisherName: Together AI
serviceCategory: AI and Machine Learning
billingModel:
  pricingCategory: Usage-Based
  billingFrequency: Monthly
  billingCurrency: USD
  chargeCategories:
  - Usage
  - Purchase
  - Adjustment
focusColumns:
  ServiceName: Together AI Cloud
  ServiceCategory: AI and Machine Learning
  ProviderName: Together AI
  PublisherName: Together AI
  InvoiceIssuerName: Together AI
  BillingCurrency: USD
  ChargeCategory: Usage
  PricingCategory: Usage-Based
meters:
- name: input_tokens
  description: Tokens sent in chat / completion requests, billed per 1M tokens per model.
  unit: tokens
  aggregation: sum
  dimensions:
  - account
  - model
  - api
- name: output_tokens
  description: Tokens generated, billed per 1M tokens per model.
  unit: tokens
  aggregation: sum
  dimensions:
  - account
  - model
  - api
- name: embedding_tokens
  description: Tokens processed for embeddings, billed per 1M tokens per model.
  unit: tokens
  aggregation: sum
  dimensions:
  - account
  - model
- name: rerank_documents
  description: Documents reranked, billed per 1K or per 1M units depending on model.
  unit: documents
  aggregation: sum
  dimensions:
  - account
  - model
- name: image_generations
  description: Image generation requests, billed per image per model/quality tier.
  unit: images
  aggregation: sum
  dimensions:
  - account
  - model
- name: video_generations
  description: Video generation requests, billed per video per model/quality tier.
  unit: videos
  aggregation: sum
  dimensions:
  - account
  - model
- name: audio_characters
  description: TTS / STT character counts, billed per 1M characters per model.
  unit: characters
  aggregation: sum
  dimensions:
  - account
  - model
- name: fine_tuning_tokens
  description: Training tokens for fine-tuning jobs, billed per 1M tokens per base-model size class.
  unit: tokens
  aggregation: sum
  dimensions:
  - account
  - base_model
  - method
- name: dedicated_endpoint_hours
  description: Wall-clock hours of dedicated inference endpoint runtime per GPU class.
  unit: hours
  aggregation: sum
  dimensions:
  - account
  - endpoint
  - gpu_class
- name: gpu_cluster_hours
  description: Wall-clock hours of bare-metal GPU cluster runtime per GPU class and reservation tier.
  unit: hours
  aggregation: sum
  dimensions:
  - account
  - cluster
  - gpu_class
  - reservation_tier
- name: batch_tokens
  description: Tokens consumed via the Batch API at up to 50% discount versus serverless.
  unit: tokens
  aggregation: sum
  dimensions:
  - account
  - model
principles:
- name: Visibility
  description: Pull token / asset / hour usage from the Together console; export invoices monthly.
- name: Allocation
  description: Tag API keys per workload/team and join usage to internal cost centers.
- name: Optimization
  description: Move non-realtime workloads to Batch (-50%); pick smaller open-source models when quality permits; reserve GPU clusters for steady-state workloads.
- name: Accountability
  description: Assign owners per endpoint, fine-tuning job, and GPU cluster reservation.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com