Same Strata patch, with benchmark table: DeepSeek V4.1 sessions 20× faster locally
sintetizacom · reddit · 2026-11-02
Duplicate of the Strata patch post with the results table: DeepSeek V4.1 agent session first token drops from 100s to 5s, fresh prompt reads from 22.8s/58GB SSD to 8.2s/1.3GB, coding suite steady at 12/13, DRAM utilization pushed from Sam's "5%" to 100%.
More from coding & agent
- Turn every LLM mistake into rolling memory by cursing at it and mining local logs — jwt0625 · 2026-11-02
- How Claude Code made a 54-second 4K60 launch film from raw screen recordings — No_Coast6186 · 2026-11-02
- Developer gets Tesla Model 3 fully integrated into Home Assistant dashboard — haydendevs · 2026-11-02
- Agents always do stupid things — and it hurts more on rented GPU time — nptacek · 2026-11-02
- NoHumans: a free shared knowledge store for agents, built almost entirely by Claude Code — dark-epiphany · 2026-11-02
- Open-source AnyShot: a Claude agent acts as cinematographer, cuts 6-hour renders to 3 minutes — Scobleizer · 2026-11-02