fix: python 3.10 typing compatibility in compact editor and claude 3 cache pricing (#38076, #38056)

[Alphaxiaoteng]

Description

This PR resolves two issues:

1. Issue [Bug]: import litellm fails on Python 3.10 — NotRequired imported from stdlib typing without fallback #38076: Python 3.10 typing compatibility in compact.py

   • In litellm/llms/anthropic/experimental_pass_through/context_management/editors/compact.py, NotRequired was imported directly from typing, which causes an ImportError on Python < 3.11.
   • Updated the import to conditionally import NotRequired from typing on Python >= 3.11 and fallback to typing_extensions on Python < 3.11.

2. Issue [Bug]: claude-3-haiku and claude-3-opus have the Sonnet 1-hour cache write price (12x over / 5x under) #38056: Claude 3 cache creation pricing above 1hr

   • Corrected cache_creation_input_token_cost_above_1hr in model_prices_and_context_window.json and model_prices_and_context_window_backup.json:
     • claude-3-haiku-20240307: 5e-07 (2x input cost 2.5e-07, was incorrectly set to 6e-06)
     • claude-3-opus-20240229: 3e-05 (2x input cost 1.5e-05, was incorrectly set to 6e-06)

Tests

• Added unit tests:
  • tests/test_litellm/llms/anthropic/experimental_pass_through/context_management/test_compact_typing_compatibility.py
  • tests/test_litellm/test_claude_3_cache_pricing.py
• Verified with pytest across Python 3.10 and Python 3.11.
• Validated model prices schema against model_prices_and_context_window.schema.json.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR adds a version-compatible NotRequired import for the compact Anthropic editor and corrects Claude 3 long-duration cache-creation prices in both model-cost maps.

• Adds compatibility and TypedDict regression tests for the compact editor.
• Updates Haiku and Opus cache pricing and tests both canonical and backup maps.
• The compatibility test does not exercise its Python 3.10 branch in the standard Python 3.12 unit workflow, and two new comments conflict with the repository’s source-comment convention.

Confidence Score: 4/5

The PR appears safe to merge, with non-blocking improvements needed in Python 3.10 regression coverage and test-comment cleanup.

The implementation and synchronized pricing updates have no established runtime defect, while the accepted findings concern an unexercised compatibility-test branch and redundant comments.

Files Needing Attention: tests/test_litellm/llms/anthropic/experimental_pass_through/context_management/test_compact_typing_compatibility.py; tests/test_litellm/test_claude_3_cache_pricing.py

Important Files Changed

Filename	Overview
litellm/llms/anthropic/experimental_pass_through/context_management/editors/compact.py	Selects NotRequired from typing on Python 3.11+ and typing_extensions on older supported runtimes; no concrete implementation defect was identified.
model_prices_and_context_window.json	Corrects the Claude 3 Haiku and Opus above-one-hour cache-creation rates in the canonical model-cost map.
litellm/model_prices_and_context_window_backup.json	Mirrors the same corrected rates in the packaged backup map.
tests/test_litellm/llms/anthropic/experimental_pass_through/context_management/test_compact_typing_compatibility.py	Adds import and TypedDict checks, but the standard Python 3.12 workflow cannot exercise the Python 3.10 fallback branch.
tests/test_litellm/test_claude_3_cache_pricing.py	Verifies both pricing maps, with two redundant comments that violate the repository’s source-comment convention.

_{Reviews (1): Last reviewed commit: "fix: python 3.10 typing compatibility in..." | Re-trigger Greptile}

[greptile-apps]

Python 3.10 Branch Untested

The test selects its assertion from the interpreter running pytest, while the standard unit workflow uses Python 3.12. As a result, the newly added typing_extensions.NotRequired fallback is not exercised there, so a Python 3.10 compatibility regression would not be caught by this test.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

Redundant Pricing Comments

This comment and the equivalent Opus comment merely restate the immediately following input_cost * 2 calculations. Removing both avoids duplicated prose that can become stale and follows the repository convention of reserving source comments for complex logic, tool directives, or TODO/FIXME items.

Context Used: CLAUDE.md (source)

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...pass_through/context_management/editors/compact.py	80.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing Alphaxiaoteng:fix/python-310-typing-and-claude-cache-pricing (2b5c4e2) with main (947dbbf)}