Estimate AWS Bedrock costs for Claude, Llama, Titan and Mistral. Enter tokens per request and volume to project daily and monthly on-demand spend.
This calculator estimates what an Amazon Bedrock workload will cost per day and per month under on-demand pricing. Pick a foundation model, enter the input and output tokens for a typical request and how many requests you expect, and it projects your token volume and spend — then lets you compare that cost across every model in the list so you can see the price gap between, say, Claude Opus and Claude Haiku for the same workload. It runs entirely in your browser; nothing you enter is sent anywhere.
Bedrock bills on-demand text inference by tokens, with separate rates for input (the prompt you send) and output (the completion the model returns). Output tokens are usually several times more expensive than input tokens, which is why a chatbot that returns long answers can cost far more than its prompt size suggests. Modelling this before you ship is the difference between a predictable bill and a surprise.
AWS quotes Bedrock on-demand rates per 1 million tokens, split into an input rate and an output rate. There are three broad pricing modes:
As a concrete anchor, Anthropic Claude 3.5 Sonnet on Bedrock is priced at roughly $3.00 per 1M input tokens and $15.00 per 1M output tokens on-demand (batch is about $1.50 / $7.50). Meta Llama and Amazon Titan/Nova models are considerably cheaper and vary by size and region. Pricing model verified against the AWS Bedrock pricing page on 12 August 2026; specific per-token rates change and vary by region, so always confirm current figures with AWS before committing a budget. The calculator ships with a built-in rate table for quick estimates; use it for relative comparison and order-of-magnitude planning rather than as a contractual quote.
Imagine a support assistant that sends 1,000 input tokens and receives 500 output tokens per request, at 5,000 requests a day over 30 days. That is 150 million input tokens and 75 million output tokens a month. On Claude 3.5 Sonnet at $3 / $15 per 1M, that is about $450 of input plus $1,125 of output — roughly $1,575 a month, with output dominating the bill despite being the smaller token count. Switching the comparison to Claude Haiku or a Llama model shows how much you would save by routing simpler requests to a cheaper model, which is the single most effective cost lever most teams have.
Because output is priced several times higher than input, the cheapest optimisation is usually to constrain response length: set a sensible max_tokens, ask for concise answers, and avoid re-sending long histories you do not need. Estimating with this calculator makes those trade-offs visible before they show up on an invoice.
On-demand text inference is billed per token, quoted per 1 million tokens, with separate input and output rates. Batch inference is cheaper for supported models, and Provisioned Throughput reserves capacity by the hour.
The tool includes a built-in rate table for quick estimates. AWS changes prices and varies them by region, so confirm current figures on the official AWS Bedrock pricing page before you commit a budget. The pricing model here was verified on 12 August 2026.
Output tokens are typically several times more expensive than input tokens. Even when a response is shorter than the prompt, it can dominate the bill.
As a rough rule, one token is about three-quarters of an English word, so 1,000 words is roughly 1,300 tokens. For a precise count, use a tokenizer.
It models on-demand per-token pricing. Provisioned Throughput is priced per hour of reserved capacity, which is a different model suited to steady, high-volume production traffic.
Yes. The comparison view prices your exact workload across every listed model so you can see the cost difference at a glance.
No. All calculation is client-side; your token counts and volumes are never transmitted.
Use the LLM token counter to measure the exact token size of a real prompt, the self-hosted LLM cost calculator to compare Bedrock against running a model on your own hardware, and the Microsoft Copilot cost calculator when you are weighing a managed AI assistant instead.
Amazon Bedrock is AWS's fully managed service for accessing foundation models from leading AI providers — including Anthropic (Claude), Meta (Llama), Amazon (Titan), Mistral, Cohere, and Stability AI. Bedrock pricing varies significantly by model, input/output token counts, and whether you use on-demand or provisioned throughput, making cost estimation essential before deploying AI workloads.
This calculator helps you estimate Bedrock costs based on your expected usage patterns, model selection, and throughput requirements — enabling informed decisions about model selection and deployment strategy.
| Pricing Model | How It Works | Best For |
|---|---|---|
| On-Demand | Pay per input/output token with no commitment | Development, testing, variable workloads |
| Batch Inference | Up to 50% discount for async processing | Large-volume offline processing |
| Provisioned Throughput | Reserved model units for guaranteed performance | Production workloads needing consistent latency |
| Model Customization | Training costs + storage + inference | Fine-tuned models for specific use cases |
| Factor | Impact on Cost |
|---|---|
| Model selection | Claude Opus vs Haiku can differ by 30-60x per token |
| Input vs output tokens | Output tokens are typically 3-5x more expensive than input |
| Context window usage | Longer prompts = more input tokens = higher cost |
| Response length | Longer outputs significantly increase per-request cost |
| Throughput needs | Provisioned throughput has a monthly minimum commitment |
| Region | Pricing varies by AWS region |