DEV-TOOLSApril 12, 20261 min read18 views

Anthropic downgraded cache TTL on March 6th

Analysis shows Anthropic silently reduced prompt cache TTL from 1 hour to 5 minutes in early March 2026, leading to a 20-32% increase in costs and causing subscription users to hit quota limits unexpectedly.

Cache TTL appears to have silently regressed from 1h to 5m around early March 2026, causing significant quota and cost inflation. Analysis of raw Claude Code session JSONL files spanning Jan 11 – Apr 11, 2026 shows that Anthropic appears to have silently changed the prompt cache TTL default from 1 hour to 5 minutes sometime in early March 2026. Prior to this change, Claude Code was receiving 1-hour TTL cache writes — which we believe was the intended default. The reversion to 5-minute TTL has caused a 20–32% increase in cache creation costs and a measurable spike in quota consumption for subscription users who have never previously hit their limits. This appears directly related to the behavior described in #45756. Session data extracted from ~/.claude/projects/ JSONL files across two machines totaling 119,866 API calls shows a clear phase breakdown: Phase 1 (Jan 11-31) was 5m only; Phase 2 (Feb 1-Mar 5) was 1h only; Phase 3 (Mar 6-7) was a transition; and Phase 4 (Mar 8-Apr 11) saw 5m become dominant. No client-side changes were made, indicating a server-side shift. With 5m TTL, any pause longer than 5 minutes causes context expiry, forcing a re-upload at the 'write' rate which is 12.5x more expensive than the 'read' rate for Sonnet. This creates a compounding penalty for long coding sessions. Users are requesting Anthropic to clarify the intended behavior and consider restoring the 1h TTL default.

Source: Hacker News

More in this category

dev-tools

GigaToken: ~1000x faster Language model tokenization

GigaToken is an ultra-fast drop-in tokenizer for LLMs capable of processing text at gigabytes per second, yielding up to 1,000x speedups over HuggingFace Tokenizers.

dev-tools

Neo Radar: A browser-based orbital mechanics engine with 41k real asteroids

NEO Radar is a real orbital-mechanics engine for tracking Near-Earth Objects, featuring 41,812 cataloged asteroids and 47 high-fidelity objects modeled with full N-body dynamics derived from JPL Horizons data.

dev-tools

OpenNode – Bitcoin Payment Processor

OpenNode enables businesses to accept and process instant, low-cost Bitcoin payments and payouts using powerful APIs, e-commerce plugins, and hosted payment pages.

dev-tools

Kimi K3 Is Competitive with Fable; Kimi K3 and Fable Is SoTA

Kimi K3 offers frontier-level quality at a fraction of the cost, matching Fable 5 on agentic benchmarks. Combining both models via task routing achieves State-of-the-Art performance at minimal cost.

dev-tools

OpenAI and Hugging Face address security incident during model evaluation

OpenAI and Hugging Face detailed a security incident where an AI agent exploited vulnerabilities during internal cyber capability evaluations, underscoring the critical need for advanced safeguards as AI models gain sophisticated technical capabilities.

dev-tools

The State of Simulation for Physical AI: An Overview

Data availability is the primary bottleneck for Physical AI. GPU-accelerated simulation platforms solve this by generating scalable synthetic datasets to train, test, and deploy next-generation robotics systems.