Anthropic Claude

Required Anthropic Claude Attributes

The following cost calculation tags are required for traces originating from Anthropic Claude API calls.

For a complete, runnable example of Claude instrumentation, see Quill — Beakpoint's reference application for multi-model LLM cost attribution. Quill's tracing.py shows the full OTel setup, and analyzer.py shows instrumented Anthropic API calls in action.

Always Required Fields

These fields are always required for Beakpoint to calculate Claude costs:

Attribute Name	Example Value	Allowed Values
gen_ai.system	anthropic	anthropic (must be this exact value)
gen_ai.request.model	claude-sonnet-4-6	Any valid Anthropic model name
gen_ai.usage.input_tokens	512	Non-negative integer
gen_ai.usage.output_tokens	128	Non-negative integer

gen_ai.system requires a SpanProcessor

The Anthropic auto-instrumentor sets gen_ai.provider.name but not gen_ai.system. Beakpoint requires gen_ai.system for pricing. Use a custom SpanProcessor to copy gen_ai.provider.name → gen_ai.system on span start. See the _GenAiSystemProcessor in Quill for the pattern, or the Track LLM Costs guide for a full walkthrough.

Optional Enrichment Attributes

These fields are optional but improve cost accuracy when provided:

Attribute Name	Example Value	Description
gen_ai.response.model	claude-sonnet-4-6-20250514	The exact model version returned in the response. When present, this takes precedence over gen_ai.request.model for pricing lookups.
gen_ai.usage.input_tokens.cache_creation	256	Tokens written into the prompt cache. Cache creation tokens are billed at a premium rate above standard input pricing.
gen_ai.usage.input_tokens.cache_read	128	Tokens read from the prompt cache. Cache read tokens are billed at a significantly reduced rate.
cloud.provider	anthropic	Provider identification. Set automatically by the _GenAiSystemProcessor pattern.

note

Claude's prompt caching uses two distinct token counters — cache_creation and cache_read — rather than the single cached counter used by OpenAI. Provide both when available to ensure accurate cost attribution.

Supported Models

Beakpoint calculates costs for the following Anthropic Claude models. Prices are per 1 million tokens (USD).

Model	Input ($/M)	Cache Creation ($/M)	Cache Read ($/M)	Output ($/M)
claude-sonnet-5	$3.00	$3.75	$0.30	$15.00
claude-sonnet-4-6	$3.00	$3.75	$0.30	$15.00
claude-sonnet-4	$3.00	$3.75	$0.30	$15.00
claude-opus-4	$15.00	$18.75	$1.50	$75.00
claude-haiku-4-5	$0.80	$1.00	$0.08	$4.00
claude-haiku-3-5	$0.80	$1.00	$0.08	$4.00

note

Prices reflect Anthropic list pricing and may change. Beakpoint keeps these rates up to date, but check the Anthropic pricing page for the latest figures.

Python Example

The quickest way to emit the required attributes is with the opentelemetry-instrumentation-anthropic package paired with a SpanProcessor that sets gen_ai.system. This example is adapted from Quill:

pip install opentelemetry-instrumentation-anthropic

from opentelemetry.instrumentation.anthropic import AnthropicInstrumentor
import anthropic

# Instrument before creating the client (see Track LLM Costs guide
# for the full TracerProvider setup including _GenAiSystemProcessor)
AnthropicInstrumentor().instrument()

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=4096,
    system="You are a helpful assistant.",
    messages=[{"role": "user", "content": "Hello!"}],
)

The instrumentation automatically sets gen_ai.provider.name, gen_ai.request.model, gen_ai.usage.input_tokens, gen_ai.usage.output_tokens, and the optional cache token counters whenever they are available in the API response. The _GenAiSystemProcessor copies gen_ai.provider.name to gen_ai.system and cloud.provider.

For full setup instructions, including how to configure the OpenTelemetry exporter for Beakpoint, see the Track LLM Costs guide.