Tip: DeepSeek Prefix-Cache Can Hit 99.8% Across 435M Tokens in a Single Day — Reasonix Proves It

GitHub·medium signal

DeepSeek-Reasonix (5.6K stars), a DeepSeek-native terminal coding agent, is engineered around prefix-cache stability as an invariant, not a feature toggle. A real user session on May 1 demonstrated 435M input tokens with a 99.82% cache hit rate, cutting costs from ~$61 to ~$12. The v0.49.0 desktop release (May 22) adds KaTeX math rendering in chat and full window/tab/session restore across relaunches. For builders running long agent sessions on DeepSeek, architecting around the byte-stable prefix cache is the single highest-leverage cost optimization.