Estimating DeepSeek V4 Pro API cost from real Codex Plus usage data - am I calculating this right?

I'm considering switching from 2× ChatGPT Plus (Codex) at $40/month to DeepSeek V4 Pro API + Cline for Angular/TypeScript enterprise development. Refactoring, new components/services, bug fixing - no subagents or git automation, just the model. I have 30 days of real token data from both accounts (total of both): Cached input: 1,451,546,240 Uncached input: 148,728,772 Output: 4,253,455 Total: 1,604,528,467 Cache hit rate: ~91% Context: Codex sends the full codebase on every turn. 54 threads, 1,671 turns, av