specification: API Commons Plans
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/Plans
provider: Fireworks AI
providerId: fireworks-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
- Multimodal
- Fine-tuning
- GPU
- Plans
description: >-
  Fireworks AI offers serverless pay-per-token inference, on-demand dedicated
  GPU deployments billed per GPU-second, batch inference at 50% of serverless,
  cached input tokens at 50% of standard, and managed fine-tuning. New users
  get $1 in free credits and postpaid billing as usage grows.
notes: >-
  Per-model rates are listed on the Fireworks pricing page; representative
  ranges are captured here. Fine-tuned model serving is the same per-token
  price as the base model.
sources:
- https://fireworks.ai/pricing
- https://docs.fireworks.ai/
plans:
- id: fireworks-serverless
  name: Serverless (Pay-as-you-go)
  type: usage
  description: >-
    On-demand per-token inference with high rate limits, postpaid billing,
    and zero cold starts.
  entries:
  - label: Chat / Vision Tokens
    name: chat_tokens
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: per 1M tokens, varies by model (see pricing page)
    userMultiplied: false
  - label: Cached Input Tokens
    name: cached_input
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: 50% of the standard input rate
    userMultiplied: false
  - label: Embeddings (up to 150M params)
    name: embeddings_small
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: $0.008 per 1M
    userMultiplied: false
  - label: Embeddings (150M-350M params)
    name: embeddings_mid
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: $0.016 per 1M
    userMultiplied: false
  - label: Embeddings (Qwen3 8B)
    name: embeddings_large
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: $0.10 per 1M
    userMultiplied: false
  elements:
  - name: Chat Completions
  - name: Vision
  - name: Embeddings
  - name: Rerank
  - name: Images
  - name: Audio
- id: fireworks-batch
  name: Batch Inference
  type: usage
  description: >-
    Asynchronous batch jobs priced at 50% of serverless input and output
    rates.
  entries:
  - label: Batch Tokens
    name: batch_tokens
    type: usage
    metric: tokens
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1000000
    price: 50% of serverless rates (input and output)
    userMultiplied: false
  elements:
  - name: Batch Chat Completions
  - name: Batch Embeddings
- id: fireworks-fine-tuning
  name: Fine-Tuning
  type: usage
  description: >-
    Supervised fine-tuning (LoRA and full) priced per 1M training tokens
    by model size and method, plus reinforcement fine-tuning billed per
    GPU-hour at on-demand deployment rates.
  entries:
  - label: Supervised Fine-Tuning Tokens
    name: ft_sft
    type: usage
    metric: tokens
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1000000
    price: $0.50-$40.00 per 1M training tokens (varies by model size and method)
    userMultiplied: false
  - label: Reinforcement Fine-Tuning
    name: ft_rft
    type: usage
    metric: hours
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1
    price: per GPU-hour at on-demand rates ($7-$12 per hour)
    userMultiplied: false
  - label: Fine-Tuned Model Serving
    name: ft_serving
    type: usage
    metric: tokens
    limit: -1
    timeFrame: month
    geo: global
    unit: 1000000
    price: same per-token price as base model
    userMultiplied: false
  elements:
  - name: SFT (LoRA)
  - name: SFT (Full)
  - name: Reinforcement Fine-Tuning
- id: fireworks-on-demand
  name: On-Demand Deployments (Dedicated GPUs)
  type: usage
  description: >-
    Pay-per-GPU-second dedicated GPU deployments with autoscaling and
    no cold-start charge.
  entries:
  - label: H100 / H200 (per hour)
    name: gpu_h100_h200
    type: usage
    metric: hours
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1
    price: $7.00 per hour
    userMultiplied: false
  - label: B200 (per hour)
    name: gpu_b200
    type: usage
    metric: hours
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1
    price: $10.00 per hour
    userMultiplied: false
  - label: B300 (per hour)
    name: gpu_b300
    type: usage
    metric: hours
    limit: -1
    timeFrame: usage
    geo: global
    unit: 1
    price: $12.00 per hour
    userMultiplied: false
  elements:
  - name: H100 / H200
  - name: B200
  - name: B300
- id: fireworks-enterprise
  name: Enterprise
  type: enterprise
  description: >-
    Reserved capacity, dedicated regions, SOC2 / HIPAA compliance support,
    and negotiated terms. Contact Fireworks sales.
  entries:
  - label: Enterprise Agreement
    name: enterprise
    type: flat
    metric: contract
    limit: -1
    timeFrame: year
    geo: global
    unit: 1
    price: contact sales
    userMultiplied: false
  elements:
  - name: Custom Volume Pricing
  - name: Reserved Capacity
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com