Teknium shares Hermes telemetry: 92.4% of tokens are cache reads
Teknium · x · 2026-11-22
Teknium (Nous Research) shared average token-mix telemetry from opt-in Hermes users. The pie chart shows cache reads account for 92.42% of all tokens, vs only 0.64% uncached input, 1.21% cache writes and 5.73% other — meaning the bulk of real-world agent token spend is cache hits on long context, with genuinely new input a tiny fraction. A rare first-party data point on agent workload cost structure and the value of prompt caching.
Related event: Teknium Reveals Over 90% of Hermes User Tokens Are Cache Reads(2 posts)→
More from coding & agent
- Tens of thousands of compute nodes power this multi-agent swarm architecture — hwchase17 · 2026-11-22
- Statechart-based actor runtime simulates 1 million foraging and fighting ants in-process — sull · 2026-11-22
- Agent ran for hours playing Smash Ultimate but botched the replay-saving task — anpaure · 2026-11-22
- Elvis Saravia opens up his "knowledge factory": a daily-automated MD knowledge graph — omarsar0 · 2026-11-22
- Team says Grok Bot agents ran half the prep for a 5,000-person summit in months — GavinSBaker · 2026-11-22
- Dev shows off browser pinball game polished with Gemini 4 Neon's help — AI_Andrew · 2026-11-22