[NingLin Liu (ninglinLiu)]

Motivation & Context

With Anthropic prompt caching, AnthropicClient set UsageDetails["input_token_count"] to Anthropic's raw usage.input_tokens, which is only the tokens after the last cache breakpoint. Total input is input_tokens + cache_creation_input_tokens + cache_read_input_tokens. Everywhere else in Agent Framework input_token_count is total input (with cache_read_input_token_count as a subset), matching OpenAI/Gemini and the GenAI semconv for gen_ai.usage.input_tokens. A cache hit therefore under-reports Claude input on spans and histograms (e.g. 12 instead of 9312).

Description & Review Guide

• What are the major changes?
  • _parse_usage_from_anthropic sums cache creation/read into input_token_count when present.
  • Cache-specific fields are unchanged.
  • Update test_parse_usage_with_cache_tokens expected total (100+20+30=150).
• What is the impact of these changes? Anthropic usage totals align with other connectors; dashboards comparing providers stop undercounting Claude when caching is used.
• What do you want reviewers to focus on? That cache fields remain separate and that streaming incremental usage still reconstructs the final total.

Related Issue

Fixes #9808

AI Assistance

• No material AI assistance was used.
• This is an AI-assisted contribution. I reviewed, understood, and verified all submitted content and accept responsibility for it.

AI assistance details: confirming Anthropic's input_tokens vs cache fields against the GenAI semconv, implementing the total in _parse_usage_from_anthropic, updating the unit test, and this PR description.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR links to an agreed issue with no competing open PR, or the Related Issue section documents a trivial-change or repository-automation exception.
• This is not a breaking change.

[Copilot]

🟡 Changes recommended

Partial streaming usage updates drop previously reported cached input tokens when cache fields are omitted, producing negative increments and undercounting total usage.

1 open issue

• Preserve cached token counts across partial streaming usage snapshots · New

What changed in this PR

This PR addresses #9808 by including Anthropic cache tokens in total input usage, aligning the Python connector with other Agent Framework providers and telemetry expectations.

Changes:

• Add cache-creation and cache-read tokens to input_token_count, preserving separate cache fields.
• Update the cached-usage unit test to expect 150 total input tokens.

File	Description
python/​packages/​anthropic/​tests/​test_anthropic_client.py	Update the expected cached-input total.
python/​packages/​anthropic/​agent_framework_anthropic/​_chat_client.py	Include cache tokens when calculating total input usage.

_{Review effort: Balanced}