Fireworks AI
## what it does
Overview
Fireworks AI is a US-based inference and training platform for open-source models. Three surfaces: Serverless (per-token, no cold starts, $1 in free credits), On-Demand deployments (per-GPU-second for prod), and Training (managed SFT/DPO/RL + a Serverless Training API for LoRA on launch models).
Key Differentiator: speed + breadth
- 100+ supported models across text, vision, audio, image, embeddings.
- OpenAI-compatible (
https://api.fireworks.ai/inference/v1) and Anthropic-compatible (/inference) endpoints. - Live routing across Standard, Priority, and Fast tiers; rate limits and pricing vary per tier — check the serverless pricing docs for current rates per model.
Pricing (fetched 2026-09-10)
Source: fireworks.ai/pricing. Per-token rates depend on model; on-demand GPUs are billed per second.
On-demand GPUs
| GPU | Price per hour (through 31 Aug 2026) | Price per hour (from 1 Sep 2026) |
|---|---|---|
| H100 80 GB | $7.00 | $8.00 |
| H200 141 GB | $7.00 | $8.00 |
| B200 180 GB | $10.00 | $13.00 |
| B300 288 GB | $12.00 | $15.00 |
| GB300 288 GB | $18.00 | $20.00 |
Region-restricted deployments are 1.5× premium. Contact Sales.
Embeddings (serverless, $/1M input tokens)
| Base model parameter count | $/1M input |
|---|---|
| up to 150M | $0.008 |
| 150M – 350M | $0.016 |
| Qwen3 8B | $0.10 |
Serverless Training API (LoRA, per 1M tokens)
| Base model | Prefill | Cached Prefill | Sample | Train |
|---|---|---|---|---|
| Qwen 3.8 27B (128K ctx) | $1.86 | $0.372 | $5.595 | $4.10 |
| Kimi K3 (192K ctx) | $10.87 | $2.17 | $27.11 | $32.55 |
| DeepSeek V4 Flash 0731 (262K ctx) | $1.74 | $0.35 | $4.33 | $5.20 |
| Muse Glimmer 30B (128K ctx) | $1.96 | $0.39 | $4.88 | $5.86 |
Checkpoint storage for serverless models is included during private preview. Frontier models beyond the launch set are coming soon.
Managed Training ($/1M training tokens)
| Size | LoRA SFT | LoRA DPO | Full SFT | Full DPO |
|---|---|---|---|---|
| up to 16B | $0.50 | $1.00 | $1.00 | $2.00 |
| 16.1B – 80B | $3.00 | $6.00 | $6.00 | $12.00 |
| 80B – 300B (e.g. Qwen3-235B, gpt-oss-120B) | $6.00 | $12.00 | $12.00 | $24.00 |
| > 300B (e.g. DeepSeek V3, Kimi K2) | $10.00 | $20.00 | $20.00 | $40.00 |
SFT / DPO prices assume non-reasoning traces; multi-turn with reasoning traces unrolls into user + assistant + thinking tokens, increasing tuned-token counts.
Quick start
export FIREWORKS_API_KEY="your_api_key"import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["FIREWORKS_API_KEY"],
base_url="https://api.fireworks.ai/inference/v1",
)
resp = client.chat.completions.create(
model="accounts/fireworks/models/deepseek-v3p1",
messages=[{"role": "user", "content": "Say hello in Spanish"}],
)
print(resp.choices[0].message.content)Reasoning models expose a reasoning_content field on the assistant message. Vision models accept image_url content parts.
Compliance
- SOC 2, HIPAA, audit reports on trust.fireworks.ai.
- US data residency; no customer data used for training.
Related
- Together AI — Alternative US inference platform
- DeepSeek Privacy Guide — Why US hosting matters
- DeepSeek — Model overview
- Serverless docs — Per-tier rate table
## embed this badge
