Fireworks AI
Fireworks AI is a high-performance inference cloud for open-source AI models, enabling developers and enterprises to build, fine-tune, and scale generative AI applications at blazing speed.
At a Glance
Get started with $1 in free credits on serverless inference. No setup required.
Engagement
Available On
Alternatives
Listed Apr 2026
About Fireworks AI
Fireworks AI is a cloud inference platform that provides fast, scalable access to open-source AI models optimized for production workloads. Founded by veterans of PyTorch, Meta, and Google, Fireworks delivers industry-leading throughput and latency across text, vision, speech, image, and embedding models. The platform supports the full AI model lifecycle — from serverless experimentation to fine-tuning and enterprise-grade on-demand GPU deployments — without requiring teams to manage infrastructure.
- Serverless Inference — Sign up and start calling models instantly with per-token pricing, no cold starts, and $1 in free credits to get started.
- Model Library — Access a broad catalog of popular open-source models including DeepSeek, Qwen, Gemma, Kimi, Llama, Mistral, FLUX, and Whisper, all optimized for cost, speed, and quality.
- Fine-Tuning — Customize open models using LoRA or full-parameter SFT, DPO, and reinforcement fine-tuning (RFT) with minimal setup; fine-tuned models are served at base model prices.
- On-Demand GPU Deployments — Reserve A100, H100, H200, B200, or B300 GPUs billed per second for higher throughput, lower latency, and higher rate limits at scale.
- Multimodal Support — Run text, vision, speech-to-text (Whisper), and image generation (FLUX, SDXL) workloads through a unified API.
- Enterprise Security & Compliance — SOC2 Type 2, HIPAA, and GDPR compliant with zero data retention, data residency support, RBAC, and SSO (Google, OIDC, SAML).
- Bring Your Own Cloud — Deploy on Fireworks' globally distributed virtual cloud or bring your own cloud environment; available via AWS and GCP marketplaces.
- Batch Inference — Run batch jobs at 50% of serverless pricing for both input and output tokens, ideal for offline or high-volume workloads.
- Observability & Reliability — Built-in failover, load balancing, auto-scaling, and comprehensive metrics dashboards for production confidence.
- Developer Tooling — OpenAI-compatible API, CLI (firectl), SDKs, cookbooks, and detailed documentation to accelerate integration.
Community Discussions
Be the first to start a conversation about Fireworks AI
Share your experience with Fireworks AI, ask questions, or help others learn from your insights.
Pricing
Serverless Free Credits
Get started with $1 in free credits on serverless inference. No setup required.
- $1 in free credits
- Per-token pricing
- High rate limits
- Postpaid billing
- Access to full model library
Serverless Inference
Pay per token for text, vision, speech, image, and embedding models with no infrastructure management.
- Text & vision models from $0.10/1M tokens
- Speech-to-text from $0.0009/audio minute
- Image generation from $0.00013/step
- Embeddings from $0.008/1M tokens
- Batch inference at 50% discount
- Cached input tokens at 50% discount
- No cold starts
- High rate limits
Fine Tuning
Supervised, preference, and reinforcement fine-tuning priced per 1M training tokens. Fine-tuned models served at base model prices.
- LoRA SFT from $0.50/1M tokens (up to 16B params)
- LoRA DPO from $1.00/1M tokens
- Full Param SFT from $1.00/1M tokens
- Full Param DPO from $2.00/1M tokens
- Reinforcement fine-tuning (RFT) priced per GPU hour
- VLM supervised fine-tuning
- Fine-tuned models served at base model prices
On-Demand GPU Deployments
Reserve dedicated GPUs billed per second for higher throughput, lower latency, and higher rate limits.
- A100 80GB GPU at $2.90/hour
- H100 80GB GPU at $6.00/hour
- H200 141GB GPU at $6.00/hour
- B200 180GB GPU at $9.00/hour
- B300 288GB GPU at $11.00/hour
- Billed per second
- No extra charges for start-up times
- Higher rate limits
- Faster speeds
Enterprise
Custom enterprise deployments with bring-your-own-cloud, compliance, SSO, RBAC, and dedicated support.
- SOC2 Type 2, HIPAA, GDPR compliant
- Bring your own cloud or run on Fireworks cloud
- Zero data retention and data sovereignty
- RBAC and SSO (Google, OIDC, SAML)
- AWS and GCP marketplace purchasing
- Dedicated Fireworks AI engineering support
- Custom rate limits and SLAs
- Observability and metrics dashboards
Capabilities
Key Features
- Serverless inference with per-token pricing
- On-demand GPU deployments (A100, H100, H200, B200, B300)
- Fine-tuning: LoRA SFT, LoRA DPO, Full Param SFT, Full Param DPO, RFT
- Support for text, vision, speech-to-text, image generation, and embeddings
- OpenAI-compatible API
- Batch inference at 50% discount
- Cached input token pricing
- Model library with 100+ open-source models
- SOC2 Type 2, HIPAA, GDPR compliance
- Zero data retention and data sovereignty
- RBAC and SSO (Google, OIDC, SAML)
- Bring your own cloud support
- Global distributed infrastructure
- Auto-scaling and load balancing
- CLI (firectl) and SDK support
- Cookbooks and documentation