The Cache TTL Trap: Why We Need Paid Retention Windows
EmetInteractive · reddit · 2026-08-30
The author discusses the cost issue of prompt cache expiring when users step away, leading to full re-ingestion costs upon return.
- Pain Point: Developers often write keep-alive scripts. However, a short break (e.g., for coffee) causes the next short prompt to re-ingest the entire context at full input price, a 10x hit for large sessions.
- Proposal: Suggests a "Retention Window" feature. Users would pin a cache key with a TTL and pay a one-time write premium (e.g., 1.25x for short-term, 2x for long-term), significantly reducing read costs for the window's duration.
- Ecosystem: Anthropic has had a 5min/hr mechanism; OpenAI recently added TTL options, and hosts like DeepInfra offer similar features. However, many wrappers hardcode 5 minutes and don't expose these controls.
More from Infra
- SK Hynix ties new US plant to looser chip equipment exports to China, sources say — pstAsiatech · 2026-10-15
- Perplexity's new embedding models: index with a 9B model, search with 0.6B — antoine_chaffin · 2026-10-15
- Unsloth ships GGUF quants for Nemotron 3.5: Bolt-74B runs in 24GB RAM — danielhanchen · 2026-10-15
- vLLM-Omni ships Day-0 serving for FLUX 3 Image Klein with BF16, FP8, NVFP4 support — vllm_project · 2026-10-15
- NVIDIA's Nemotron 3.5 Super VL lands training and inference support on Applied Compute — ypatil125 · 2026-10-15
- CoreWeave bond yield tops 16% as loans outlast customer contracts by ~2 years — tengyanAI · 2026-10-15