### Check for existing issues

I searched the existing issues and did not find a duplicate.

### What happened?

`input_cost_per_token_cache_hit` is declared in the pricing schema (`ModelInfoBase` in [`litellm/types/utils.py`](https://github.com/BerriAI/litellm/blob/main/litellm/types/utils.py)) and accepted by the field-name allowlist in `tests/test_litellm/test_utils.py`, but **no code path in the cost calculator reads it**. The canonical key actually consumed by the calculator is `cache_read_input_token_cost`.

When a model entry declares only the legacy key:

- `cache_read_input_token_cost` resolves to `None` → `0.0` via `_get_cost_per_unit` - Prompt-cache-hit tokens are subtracted from `text_tokens` in [`_calculate_input_cost`](https://github.com/BerriAI/litellm/blob/main/litellm/litellm_core_utils/llm_cost_calc/utils.py) (so they are *not* charged at the full input rate either) - Net effect: cache-hit tokens are billed at **$0** rather than the intended cache-read rate

This is a footgun for both human contributors and AI-assisted PRs adding new model entries: copying nearby DeepSeek configs that historically used `input_cost_per_token_cache_hit` produces a config that **look…