AWS Bedrock Pricing 2026:
Claude, Llama, Titan & All Models

AWS Bedrock Model Pricing (2026)

AWS Bedrock vs Direct API: Which Is Cheaper?

For most models, AWS Bedrock charges a small premium over direct API access for the convenience of AWS integration:

• Claude Sonnet via Anthropic API: $3.00/M input, $15.00/M output
• Claude Sonnet via AWS Bedrock: $3.00/M input, $15.00/M output (same price)
• Llama 3.1 70B via Groq: $0.59/M input, $0.79/M output
• Llama 3.1 70B via AWS Bedrock: $2.65/M input, $3.50/M output (4× more expensive)

Key insight: AWS Bedrock's value is not price — it's AWS ecosystem integration (IAM, VPC, CloudWatch, compliance). For pure cost, direct APIs are usually cheaper.

AWS Bedrock Provisioned Throughput

For predictable, high-volume workloads, Bedrock offers Provisioned Throughput (PT) — reserved model units billed hourly:

• Minimum commitment: 1 month
• Benefit: Guaranteed throughput, lower per-token cost at high volume
• Break-even: Typically at 50M+ tokens/month
• Example: Claude Haiku PT at 100M tokens/month saves ~30–40% vs on-demand

AWS Bedrock Batch Inference (50% Off)

Like OpenAI's Batch API, Bedrock offers batch inference at 50% discount for asynchronous workloads:

• Submit S3-stored JSONL files
• Results written back to S3
• Typical completion: 6–24 hours
• Supported models: Claude, Llama, Titan, Mistral

Why Choose AWS Bedrock?

• ✅ AWS account consolidation — single bill, existing credits, enterprise agreements
• ✅ VPC endpoint support — traffic stays in your AWS network, never public internet
• ✅ IAM-based access control — no separate API key management
• ✅ CloudWatch monitoring — integrated with your existing observability stack
• ✅ HIPAA/SOC2/GDPR compliance — inherited from AWS infrastructure
• ✅ Model diversity — switch between providers without new contracts
• ❌ Not cheapest for Llama — Groq/Together are 4× cheaper for Llama models
• ❌ No ChatGPT/GPT-4o — OpenAI not available on Bedrock