Best LLM Inference Providers

Vendor profileChecked Nov 8, 2026

Fireworks AI

Fireworks AI, Inc. · fireworks.ai · Last checked:

  • Serverless per-token API
  • Dedicated GPU deployments (on-demand)
  • Bring-your-own-model uploads

Company status

Active, privately held.

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Capabilities

  • Serverless open-weight models billed per token with Standard, Priority, Fast, Batch, and Spot modes
  • On-demand GPU deployments billed by GPU time; idle deployments scale to zero after one hour by default
  • Upload custom base models (Hugging Face format) and LoRA adapters to on-demand deployments
  • OpenAI client library compatibility (documented for the OpenAI Python client)
  • Zero data retention by default for open models: prompts and generations not logged or stored without opt-in (Response API store=True keeps conversation data up to 30 days)
  • Fine-tuning: managed training, serverless training API, dedicated training API

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Who it tends to fit

  • Teams wanting serverless and dedicated deployments for open-weight and custom models
  • Regulated teams that need HIPAA with a BAA and published ISO certifications

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Pricing

Published per-token and GPU-hour pricing on https://fireworks.ai/pricing (for example, H100 80 GB listed at $8.00 per hour for on-demand deployments). Serverless usage draws down prepaid credits.

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Published certifications

  • SOC 2 Type II
  • HIPAA (with a BAA)
  • GDPR
  • ISO 27001:2022
  • ISO 27701
  • ISO/IEC 42001:2023

Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Limitations to check

  • Response API stores conversation data by default (store=True) for up to 30 days unless you set store=False
  • Custom base models and uploaded LoRA adapters run only on on-demand deployments, not serverless
  • GPU availability varies by region

Sources 1, 2, 3, 4, 5 · checked 2026-11-08

Sources

  1. Pricing: serverless per-token modes (Standard, Priority, Fast, Batch, Spot); on-demand GPU per-hour (H100 $8.00/hr); SOC 2 Type II, HIPAA with BAA, zero data retention by defaultfireworks.ai/pricingChecked 2026-11-08
  2. Trust Center lists SOC 2 Type 2, HIPAA, GDPR, ISO 27001:2022, ISO 27701, ISO/IEC 42001:2023trust.fireworks.aiChecked 2026-11-08
  3. No logging or storage of prompt or generation data for open models without opt-in; Response API store=True default, up to 30 daysdocs.fireworks.ai/guides/security_compliance/data_handlingChecked 2026-11-08
  4. OpenAI Python client library can be used with Fireworksdocs.fireworks.ai/tools-sdks/openai-compatibilityChecked 2026-11-08
  5. Custom base model upload in Hugging Face format; on-demand deployments onlydocs.fireworks.ai/models/uploading-custom-modelsChecked 2026-11-08

Report a correction: [email protected]. Facts are re-checked at least every 90 days.