## Problem

Developers integrating litellm into a production app need a quick way to estimate the dollar cost of a prompt before deploying it. Today `litellm.completion_cost()` is a public function that computes a cost from a `ModelResponse` (or from `model`, `messages`, `prompt`, `completion` directly), but it lives in the Python API and isn't easily runnable from a shell. To estimate the cost of "this prompt on `gpt-4o`", a developer has to open a Python REPL, import the right symbols, and call the function. There is no `litellm cost-estimate` or similar CLI in the existing `litellm/proxy/client/cli/commands/` (which only manages a running proxy), and no issue tracking this gap.

## Motivation

Pre-deployment cost estimation is a real production concern. A team adopting litellm for the first time typically wants to know: - "What will this prompt cost on `gpt-4o`?" - "How much more expensive is `claude-opus-4-8` than `claude-sonnet-4-6` for our traffic shape?" - "If we move to `gemini-2.5-pro` for the cache-heavy traffic, what's the rough cost?"

Today, the only ways to answer those questions are: (1) open a Python REPL, (2) read the `model_prices_and_context_window.json` manually…