BedrockExplorer

Llama 3.3 70B Instruct

Meta · Active · available in 3 of 31 regions

Cost calculator Compare
Tool useBatch inferenceStreaming VisionPrompt cachingEmbeddingsFine-tuning
Context window
128K
128,000 tokens
Max output
—
Input / 1M
$0.720
us-east-1
Output / 1M
$0.720

What Llama 3.3 70B Instruct is good for

editorial

Llama 3.3 70B Instruct offers a 128K-token context window with tool use at $0.720 per 1M input tokens. Callable in 3 of 31 regions.

Suits

  • Tool use. Can call functions, so it can drive retrieval, look things up, and take actions rather than only answering.
  • Bulk jobs. Batch inference is available, typically at half the on-demand rate, for work that can wait — backfills, nightly enrichment, evaluation runs.

Think twice if

  • Agentic loops. No prompt caching published, so every step re-pays full price for the prompt prefix. Expensive for agents that resend a long context.
  • Documents with layout. Text only. Scanned PDFs, screenshots and charts need a vision model or an OCR step first.
  • Data residency. Available in only 3 of 31 regions, so it may not clear a residency requirement.

This section is judgement, not data from AWS. It is composed from the capabilities, context window, price position and region coverage shown elsewhere on this page — so it stays in step with the daily snapshot rather than going stale.

Where teams typically use it

Chat and assistants

Conversational work where responsiveness matters as much as depth.

  • · A customer-facing assistant where the first token needs to appear in under a second.
  • · An in-product copilot that explains what the user is looking at and answers follow-ups.
  • · A triage bot that qualifies an incoming request before routing it to the right team.

Coding

Writing, reviewing and refactoring code, usually across more than one file.

  • · A review pass that flags real defects with a severity and leaves style alone.
  • · A framework upgrade applied across a repository, one module at a time, tests green between each.
  • · Turning a failing bug report into a reproducing test and then a fix.

Model & inference profile IDs

Base model ID — direct on-demand invoke

Cross-region inference profiles (1)

Regions marked Inference profile only require a profile ID rather than the base model ID — invoking the base ID there returns a validation error. Regional (us., eu.) profiles carry a 10% premium over global.

Pricing by region

USD per 1M tokens
Region Input Output Cache read Cache write Batch in Batch out Source
us-east-1 US East (N. Virginia) $0.720 $0.720 — — $0.360 $0.360 API
us-east-2 US East (Ohio) $0.720 $0.720 — — $0.360 $0.360 API
us-west-2 US West (Oregon) $0.720 $0.720 — — $0.360 $0.360 API

Region availability

Inference profile only us-east-1 Profile only On-demand us-east-2 On-demand Not available us-west-1 Not available Inference profile only us-west-2 Profile only Not available ca-central-1 Not available Not available ca-west-1 Not available Not available mx-central-1 Not available Not available eu-west-1 Not available Not available eu-west-2 Not available Not available eu-west-3 Not available Not available eu-central-1 Not available Not available eu-central-2 Not available Not available eu-north-1 Not available Not available eu-south-1 Not available Not available eu-south-2 Not available Not available ap-east-2 Not available Not available ap-northeast-1 Not available Not available ap-northeast-2 Not available Not available ap-northeast-3 Not available Not available ap-south-1 Not available Not available ap-south-2 Not available Not available ap-southeast-1 Not available Not available ap-southeast-2 Not available Not available ap-southeast-3 Not available Not available ap-southeast-4 Not available Not available ap-southeast-5 Not available Not available ap-southeast-6 Not available Not available ap-southeast-7 Not available Not available il-central-1 Not available Not available af-south-1 Not available Not available sa-east-1 Not available

TPM / RPM are default account quotas from AWS Service Quotas where a model-specific limit is published. They are per-account defaults and adjustable on request.

Where this model actually runs

Under a strict residency constraint, Llama 3.3 70B Instruct can be served without the request leaving 1 of the 25 jurisdictions with a Bedrock region.

United States All jurisdictions →

Llama 3.3 70B Instruct appears in

Tool use / function calling Batch inference

Common questions

How much does Llama 3.3 70B Instruct cost on Amazon Bedrock?
$0.720 per 1M input tokens and $0.720 per 1M output tokens in us-east-1.
Which AWS regions support Llama 3.3 70B Instruct?
3 of 31 regions: us-east-1, us-east-2, us-west-2. Regions marked profile-only need a cross-region inference profile ID such as us.meta.llama3-3-70b-instruct-v1:0 rather than the bare model ID.
What is the context window of Llama 3.3 70B Instruct?
128,000 tokens (128K).
What is the model ID for Llama 3.3 70B Instruct on Bedrock?
meta.llama3-3-70b-instruct-v1:0. In regions where it is only reachable through a cross-region inference profile, use a prefixed ID instead: us.meta.llama3-3-70b-instruct-v1:0.

Other Meta models

Llama 3 70B Instruct Llama 3 8B Instruct Llama 3.1 70B Instruct Llama 3.1 8B Instruct Llama 4 Maverick 17B Instruct Compare these →