# Fireworks AI API pricing and models

Fireworks AI publishes per-token prices for 18 text models, as of October 20, 2026. The lowest output rate is $0.20 per million tokens, on NVIDIA Nemotron 3.5 Lightning 30B A3B, and the highest-scoring model it serves is Qwen3.8 Max, at 81 on our 100-point score.

Fireworks AI serves open-weight models on shared serverless endpoints and on dedicated deployments. This page covers the serverless prices.

Source: [Fireworks serverless pricing docs (linked from fireworks.ai/pricing)](https://docs.fireworks.ai/serverless/pricing), read 2026-10-20. Prices in US dollars per million tokens.

## Models we score

| Model | Score | Input | Output | Cached input | Context | vs maker |
|---|---|---|---|---|---|---|
| Qwen 3.8 Max (`accounts/fireworks/models/qwen3p8-max`) | 81 | $2.00 | $6.00 | $0.25 | — | — |
| Kimi K3 (`accounts/fireworks/models/kimi-k3`) | 79 | $3.00 | $15.00 | $0.30 | — | Same |
| DeepSeek V4.1 Flash (`accounts/fireworks/models/deepseek-v4p1-flash`) | 73 | $0.30 | $1.20 | $0.006 | — | Same |
| DeepSeek V4.1 Flash 1017 (`accounts/fireworks/models/deepseek-v4p1-flash-1017`) | 73 | $0.30 | $1.20 | $0.006 | — | Same |
| Muse Spark Open (`accounts/fireworks/models/muse-spark-open`) | 72 | $1.25 | $4.25 | $0.15 | — | — |
| GLM 5.3 Flash (`accounts/fireworks/models/glm-5p3-flash`) | 69 | $0.15 | $0.50 | $0.03 | — | — |
| GLM 5.3 (`accounts/fireworks/models/glm-5p3`) | 68 | $1.40 | $4.40 | $0.26 | — | — |
| MiniMax M3 (`accounts/fireworks/models/minimax-m3`) | 65 | $0.30 | $1.20 | $0.06 | — | Same |
| NVIDIA Nemotron 3 Ultra (Preview) (`accounts/fireworks/models/nemotron-3-ultra-nvfp4`) | 57 | $0.60 | $2.40 | $0.12 | — | — |
| OpenAI GPT OSS 120B (`accounts/fireworks/models/gpt-oss-120b`) | 55 | $0.15 | $0.60 | $0.015 | — | — |
| Ember-1 (`accounts/fireworks/models/ember-1`) | 0 | $3.00 | $15.00 | $0.30 | — | Same |

## Other models

| Model | Input | Output | Cached input | Context |
|---|---|---|---|---|
| NVIDIA Nemotron 3.5 Lightning 30B A3B (`accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b`) | $0.05 | $0.20 | $0.01 | — |
| GLM 5.3 Flash (US) (`accounts/fireworks/models/glm-5p3-flash`) | $0.23 | $0.75 | $0.045 | — |
| DeepSeek V4.1 Flash (US) (`accounts/fireworks/models/deepseek-v4p1-flash`) | $0.45 | $1.80 | $0.009 | — |
| GLM 5.3 (US) (`accounts/fireworks/models/glm-5p3`) | $2.10 | $6.60 | $0.39 | — |
| GLM 5.3 Fast (`accounts/fireworks/models/glm-5p3`) | $2.10 | $6.60 | $0.39 | — |
| Kimi K3 (US) (`accounts/fireworks/models/kimi-k3`) | $4.50 | $22.50 | $0.45 | — |
| Kimi K3 Fast (`accounts/fireworks/models/kimi-k3`) | $4.50 | $22.50 | $0.45 | — |

## Models without a named price

- Less than 4B parameters: $0.10 in, $0.10 out
- 4B-16B parameters: $0.20 in, $0.20 out
- More than 16B parameters: $0.90 in, $0.90 out
- MoE up to 56B parameters (e.g. Mixtral 8x7B): $0.50 in, $0.50 out
- MoE 56.1B-176B parameters (e.g. DBRX, Mixtral 8x22B): $1.20 in, $1.20 out

## Billing notes

- The table shows Standard rates. Priority mode costs about 1.25 to 1.5 times more, and the US-only variants cost 1.5 times the Standard rate.
- Batch jobs cost half the serverless rate on input and output. Models without a named price are billed by parameter-count tier, listed below the table.

## What we measured (2026-09-01)

Fireworks invoiced the highest cost per DeepSeek Flash token in the run, $0.23 per million blended, 5.3 times OpenRouter. Among the platforms that itemize caching it cached the least. Its GLM-5.2 needle-hit rate was 0.83; every other platform scored 0.91 or better on the same weights.

| Model run | Invoiced /1M blended | Median latency | p95 latency | Succeeded |
|---|---|---|---|---|
| DeepSeek V4 Flash | $0.234 | 2.72 s | 8.28 s | 98.7% |
| GLM-5.2 | $1.338 | 5.94 s | 8.47 s | 99.8% |

Method and invoices: https://benchlm.ai/openrouter-alternatives

## Hosts that sell the same models

- [OpenRouter](https://benchlm.ai/hosts/openrouter): 9 shared, 4 cheaper per output token (DeepSeek V4.1 Flash 1017: $0.12 vs $1.20)
- [DeepInfra](https://benchlm.ai/hosts/deepinfra): 7 shared, 6 cheaper per output token (DeepSeek V4.1 Flash 1017: $0.29 vs $1.20)
- [Novita AI](https://benchlm.ai/hosts/novita): 7 shared, 1 cheaper per output token (GPT-OSS 120B: $0.25 vs $0.60)
- [Nebius Token Factory](https://benchlm.ai/hosts/nebius): 7 shared, 0 cheaper per output token
- [Baseten](https://benchlm.ai/hosts/baseten): 6 shared, 1 cheaper per output token (GPT-OSS 120B: $0.50 vs $0.60)
- [Together AI](https://benchlm.ai/hosts/together): 6 shared, 0 cheaper per output token

## FAQ

### How much does the Fireworks AI API cost?

Fireworks AI bills per million tokens, with separate input and output rates for each model. On October 20, 2026 its output rates ran from $0.20 on NVIDIA Nemotron 3.5 Lightning 30B A3B to $22.50 across 18 paid models. Input is usually cheaper than output, so a long prompt with a short reply costs less than the output rate suggests.

### Which models can you run on Fireworks AI?

18 text models are listed, and 10 of them match a model we score. The highest-scoring is Qwen3.8 Max at 81. The table on this page has every model with its input, output and cached-input rate, the context window, and the date we read the price.

### Is Fireworks AI fast?

In our paid September 1, 2026 run, Fireworks AI returned a median of 2.72 seconds on DeepSeek V4 Flash and 5.94 seconds on GLM-5.2, over 450 requests per model. That is one workload on one date, so check your own prompts before you rely on it.

### What are the alternatives to Fireworks AI?

OpenRouter lists 9 of the same models and charges less output per token on 4 of them. DeepInfra and Novita AI also overlap. Price is only part of the choice: speed, rate limits and which model snapshot a host serves differ too.

Canonical page: https://benchlm.ai/hosts/fireworks