Show HN: Does per-step reasoning effort save money in Claude Code?

AI Digest

文章测试了Claude Code中按步骤调整推理努力对成本的影响，结果显示最大努力下节省55%费用，且不重置缓存，通过动态优化资源分配实现成本控制。

Testing per-step reasoning effort adjustments in Claude Code showed 55% cost savings at max effort without cache reset, enabling dynamic resource optimization.

Key points

• 按步骤调整推理努力在最大努力下节省55%成本，保持测试通过 Max effort per-step adjustment saved 55% cost while maintaining test passes
• 改变努力不重置缓存，节省重新处理上下文的费用 Effort changes retain cache, avoiding reprocessing context costs
• Jev模型在每一步选择不同努力级别，优化资源分配 Jev model dynamically selects effort levels per step
• 缓存命中率高达99.1%，减少重复计算 99.1% cache hit rate reduces redundant computations
• Claude Code的缓存机制允许动态调整努力，不影响历史上下文 Claude Code's caching preserves historical context with dynamic effort

Takeaway: 按步骤调整推理努力可显著降低Claude Code的计算成本。 / Per-step reasoning effort adjustment significantly reduces Claude Code's computational costs.

Why it matters 提供实际案例，展示动态调整推理资源如何优化AI编码成本，对开发者有实操参考价值。