AI & Machine Learning•September 21, 2026
Anthropic Slashes API Latency by 85% with Cross-Agent Ephemeral Cache Sharing
Anthropic has rolled out CacheMesh v3 for Claude models, enabling autonomous agent swarms to share multi-megabyte system contexts at near-zero incremental token pricing.
Official Press Release
AnthropicClaudePrompt CachingAPI CostsAI Agents
Anthropic today announced an update to its prompt caching protocol across the Claude 3.5 and Claude 4 model families, allowing distributed autonomous agents to share pre-computed KV-cache states in real time.
The update slashes token latency from seconds to under 200 milliseconds for complex codebases and enterprise knowledge bases exceeding 500,000 tokens, significantly reducing the cost barriers associated with multi-agent orchestration.
Subscribe for Updates
Get official press announcements and version releases sent directly to your email.