inference/research
verifiedmedium · 0.70updated 2026-10-16

Amazon Bedrock managed inference

aws-bedrock

Fields

id
services/svc-aws-bedrock
Provider
aws-bedrock
Endpoint
https://bedrock-runtime.us-east-1.amazonaws.com
Price basis
USD per 1M TOKENS, PER-MODEL PER-REGION list table - this is why both price fields are null: Bedrock has no single price. Every model is priced separately per AWS region via provider tabs on aws.amazon.com/bedrock/pricing (Anthropic, Meta, DeepSeek, OpenAI, Qwen, Mistral, xAI, Z AI and others), the price cells render client-side and did not extract on 2026-10-14, and regional prices differ (a Meta or Anthropic model in us-east-1 is not the eu-west-1 number). Known structural facts, first-party: batch inference = 50% below on-demand for select FMs; service tiers are Standard / Flex / Priority / Reserved; provisioned throughput is a separate per-model-unit hourly price. One anchor row visible in the legacy table: Llama 2 Chat 70B $1.95/$2.56.
Regions
us-east-1us-west-2eu-west-1eu-central-1ap-southeast-1ap-northeast-1
notes
2026-10-16 check: Artificial Analysis's Llama 3.3 70B provider table now includes an Amazon Bedrock row (context 128k, function calling linked) but the speed, latency and price cells all rendered blank, so measured_* stay null and the price stays null — a blank cell is not a zero. The 2026-10-14 basis stands: Bedrock prices per model per region and the pricing page did not extract. The compliance-and-regions answer rather than the cheap answer: Bedrock's value is that models run inside the buyer's AWS account boundary with IAM, CloudTrail and PrivateLink, across ~30 regions - Artificial Analysis benchmarks it on Llama 3.3 70B alongside the independent providers but the live cells did not extract today, so the measured fields stay null. Prices are quoted per model + region + tier; cross-region inference profiles (US/EU/geo CRIS) add another pricing axis. models_served is a sample of the FM catalog spelled as repo model slugs; provider-side ids are namespaced (anthropic.claude-*, meta.llama*).

Notes

2026-10-16 check: Artificial Analysis's Llama 3.3 70B provider table now includes an Amazon Bedrock row (context 128k, function calling linked) but the speed, latency and price cells all rendered blank, so measured_* stay null and the price stays null — a blank cell is not a zero. The 2026-10-14 basis stands: Bedrock prices per model per region and the pricing page did not extract. The compliance-and-regions answer rather than the cheap answer: Bedrock's value is that models run inside the buyer's AWS account boundary with IAM, CloudTrail and PrivateLink, across ~30 regions - Artificial Analysis benchmarks it on Llama 3.3 70B alongside the independent providers but the live cells did not extract today, so the measured fields stay null. Prices are quoted per model + region + tier; cross-region inference profiles (US/EU/geo CRIS) add another pricing axis. models_served is a sample of the FM catalog spelled as repo model slugs; provider-side ids are namespaced (anthropic.claude-*, meta.llama*).

Cross-references

Models (9)