fix(models): take the final input and cache counts from Anthropic message_delta

[pcbeingused333]

Link to Issue or Description of Change

2. Or, if no issue exists, describe the change:

Problem:

The streaming path of AnthropicLlm reads the input and cache token counts from message_start only and keeps them to the end. message_delta also reports input_tokens, cache_read_input_tokens and cache_creation_input_tokens, cumulative and final, and they were ignored. When message_start could not know the final values, a streamed turn reported fewer prompt tokens than the same turn without streaming.

Solution:

While streaming, keep the uncached input, cache-read and cache-write counts apart, let message_delta replace each count it carries (a field the event leaves out keeps its message_start value), and sum them once into prompt_token_count when the final response is built. When message_delta has no input or cache fields the result is the same as before.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Added test_streaming_takes_final_input_and_cache_counts_from_message_delta: message_start reports 3 input tokens and no cache, message_delta reports 120 input, 30 cache-read and 10 cache-write tokens, and the final response must report a prompt count of 160, 30 cached, 10 cache-creation and a total of 167. pytest tests/unittests/models/test_anthropic_llm.py: 184 passed (anthropic 1.15.0). On main only the new test fails. pyink --check and isort --check are clean on the touched files.

Manual End-to-End (E2E) Tests:

Not run against the real API; the stream events in the test follow the shape of the Anthropic streaming documentation.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

This change was drafted with Claude Code; the tests above were run before opening the PR.

🤖 Generated with Claude Code