Overview
Fireworks AI is an inference platform for open-weight models. Its serverless API bills per token for models such as DeepSeek V4.1 Flash, GLM 5.3, Qwen 3.8 and NVIDIA Nemotron, and the same account offers on-demand GPU deployments and fine-tuning. Serverless models come in Standard, Priority and Fast tiers, so you can pay more for lower latency on the same model.
Prices on the Standard tier start very low (Nemotron 3.5 Lightning 30B at $0.05 in and $0.20 out per million tokens) and include a separate cached-input rate. Batch inference is billed at 50% of serverless prices on both input and output.
New accounts get $1 in free credits, but without a payment method the API is capped at 10 requests per minute. Postpaid accounts move through spend tiers that cap monthly spend at $50, $500, $5,000 and $50,000 as cumulative spend grows, which also protects against runaway bills.
Pricing and plans
| Plan | Price | What you get |
|---|---|---|
| NVIDIA Nemotron 3.5 Lightning 30B | $0.05 in / $0.20 out per 1M | Cached input $0.01 |
| GLM 5.3 Flash | $0.15 in / $0.50 out per 1M | Cached input $0.03 |
| DeepSeek V4.1 Flash | $0.30 in / $1.20 out per 1M | Cached input $0.006 |
| Qwen 3.8 Max | $2 in / $6 out per 1M | Cached input $0.25 |
| Batch inference | 50% off | Of serverless input and output prices |
| Embeddings | From $0.008 per 1M | Models up to 150M parameters |
| Spend tiers | $50 to $50,000/mo | Tier 1 with a payment method $50; Tier 2 after $50 $500; Tier 3 after $500 $5,000; Tier 4 after $5,000 $50,000 |
What it does
- Serving tiers
- Standard, Priority and Fast pricing for the same model to trade cost against latency.
- Batch inference
- 50% of serverless pricing on both input and output.
- Cached input pricing
- Separate cached-input rate per model, e.g. $0.006 per 1M on DeepSeek V4.1 Flash.
- Spend tiers
- Monthly spend ceilings from $50 to $50,000 that rise with cumulative spend.
- On-demand deployments
- Dedicated GPU deployments and fine-tuning alongside serverless.
Strengths and limitations
Strengths
- Very low entry prices on small open models.
- Monthly spend ceilings on postpaid accounts limit surprise bills.
- Per-model choice of serving speed tier.
Limitations
- Only $1 of free credit, and 10 RPM until a payment method is added.
- New postpaid accounts are capped at $50 a month until they spend more.
- Full model pricing lives in the docs rather than the main pricing page.
- No proprietary frontier models.
Who it suits
Good for
- Production apps on open-weight models that need predictable latency tiers
- Offline batch jobs at half price
Look elsewhere if
- Free, long-running hobby use
- Teams wanting OpenAI, Anthropic or Google models
Alternatives to Fireworks AI
| Tool | From | Free option | Stages |
|---|---|---|---|
| Cerebras Inference Inference provider (open-weight models) | Usage-based | ||
| Mistral API (Mistral Studio) AI model API (model maker) | Usage-based | Free tier | |
| Claude API AI model API (model maker) | Usage-based | ||
| OpenRouter Model router (unified API) | — | Free tier | |
| OpenAI API AI model API (model maker) | Usage-based | ||
| Together AI Inference provider (open-weight models) | Usage-based |
Questions
Does Fireworks AI have free credits?
The pricing page says new accounts get started with $1 in free credits; without a payment method the API is limited to 10 requests per minute.
How much is batch inference on Fireworks?
Batch inference is billed at 50% of serverless pricing on both input and output tokens.
What is the Fireworks spend limit for a new account?
Tier 1, reached by adding a valid payment method, has a $50 monthly limit; spending or adding $50 moves you to Tier 2 with $500.
Sources read for this page