LobsterAI report: DeepSeek cache hit rate hits 97.3%, cutting effective cost to ~6.2% of list price
lxfater · x · 2026-10-19
A thread breaks down LobsterAI's September China Office Agent User Behavior Report:
- The most common model pairing is the DeepSeek Flash series + GLM 5.3 Flash (14.0%), followed by DeepSeek Flash + V4 Pro (12.4%).
- Cache economics matter for agents, which repeatedly re-read context: DeepSeek V4.1 Flash hit a 97.3% cache hit rate, bringing effective cost to 6.2% of list price (V4 Flash 6.2%, V4 Pro 7.5%). Great for iterative coding work.
- Zhipu GLM's token share on LobsterAI grew 166.7% MoM in September, while DeepSeek (-0.3%), Alibaba Qwen (-18.2%), MiniMax (-66.7%) and Kimi (-78.6%) declined.
- During the National Day holiday, general coding tasks rose the most, and agent usage peak shifted from 4pm to 10pm — people are vibe coding at night.
Related event: LobsterAI Report: Office Agent DAU Soars 83% as Task Scale Triples(4 posts)→
More from coding & agent
- 'Prayer-based programming' emerges as AI coding's latest self-deprecating meme — lxfater · 2026-10-19
- Developer: I keep playing 'human API' in my AI workflows — jacalulu · 2026-10-19
- Vibe coding a parent's TV: Claude Code strips 25 Xiaomi packages, kills boot ads — HarperSCarroll · 2026-10-19
- Sentry co-founder's AI writing workflow: dictate ideas, let an agent outline, write it yourself — lmoroney · 2026-10-19
- CarryOn: object-oriented context runtime for long-horizon agents boosts OSWorld by 5.8% — McAuley-Lab · 2026-10-19
- TreeOPSD: Tencent and Tsinghua's critic-free RL for long-horizon LLM agents — Tencent-Hunyuan · 2026-10-19