AWS Bedrock vs Fireworks AI
Same workloads, both price lists, refreshed daily. On shared line items today: AWS Bedrock is cheaper on 6 of 18 shared models (input or output price), Fireworks AI on 3.
AWS Bedrock
Claude, Llama, Mistral, Nova and more inside AWS with IAM, VPC and compliance controls. The default when your stack already lives on AWS.
- Enterprise IAM/VPC integration
- Multiple model families
- Provisioned throughput options
Fireworks AI
Speed-focused open-model inference with FireAttention serving stack, function-calling models and image generation. Strong latency SLAs for production apps.
- Very low latency serving
- Function-calling optimized
- Enterprise SLAs
Shared models, priced by both
List prices refreshed 2026-08-23 ยท input $/1M tokens| Model | AWS Bedrock in/out | Fireworks AI in/out | Cheaper input |
|---|---|---|---|
| GPT-OSS-Safeguard-20b | $0.070 / $0.200 | $0.500 / $0.500 | AWS Bedrock |
| GPT-OSS 20B | $0.075 / $0.300 | $0.070 / $0.300 | Fireworks AI |
| Ministral-3-3b-2512 | $0.100 / $0.100 | $0.100 / $0.100 | tie |
| GPT-OSS 120B | $0.150 / $0.600 | $0.150 / $0.600 | tie |
| Mistral-7b | $0.150 / $0.200 | $0.200 / $0.200 | AWS Bedrock |
| Ministral-3-8b-2512 | $0.150 / $0.150 | $0.200 / $0.200 | AWS Bedrock |
| GPT-OSS-Safeguard-120b | $0.150 / $0.600 | $1.20 / $1.20 | AWS Bedrock |
| Ministral-3-14b-2512 | $0.200 / $0.200 | $0.200 / $0.200 | tie |
| Gemma 3 27B | $0.230 / $0.380 | $0.900 / $0.900 | AWS Bedrock |
| MiniMax M2.1 | $0.300 / $1.20 | $0.300 / $1.20 | tie |
| MiniMax M2 | $0.300 / $1.20 | $0.300 / $1.20 | tie |
| Mixtral 8x7B | $0.450 / $0.700 | $0.500 / $0.500 | AWS Bedrock |
| DeepSeek V3 | $0.580 / $1.68 | $0.900 / $0.900 | AWS Bedrock |
| GLM-4.7 | $0.600 / $2.20 | $0.600 / $2.20 | tie |
| Kimi K2 Thinking | $0.600 / $2.50 | $0.600 / $2.50 | tie |
| Kimi K2.5 | $0.600 / $3.03 | $0.600 / $3.00 | tie |
| DeepSeek V3.2 | $0.620 / $1.85 | $0.560 / $1.68 | Fireworks AI |
| DeepSeek R1 | $1.35 / $5.40 | $3.00 / $8.00 | AWS Bedrock |
Cheaper-on-count is a tally of listed prices, not a quality verdict: throughput, reliability and quantization differ between providers. Disclosure.