Redditor's 3-Day SoL-Pi Test: Memory Objects Save ~12k Tokens Per Tool Run
Garblyx · reddit · 2026-09-17
Reddit user Garblyx shares a 3-day review of SoL-Pi on long-horizon tasks with multi-turn tool calling. Key takeaway: value only shows once memory objects accumulate — a combined tool run plus memory recall saved roughly 12k tokens.
- Runs drowzeys/keys-DeepSeekV4-Flash-GA-0731-Dspark (abliterated/anchored) on a modified MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark setup
- Used for software reverse engineering and red-teaming against students
- Workflow: distinct agent profiles per task, each scoped tightly to a few concrete deliverables; no perceived performance loss
- Verdict: 10/10
More from coding & agent
- The Superdark Factory: by 2029, code review becomes a formality as agents write 35T tokens a month — bratton · 2026-09-17
- He runs his entire company on OpenClaw: an early look at agent-native operations — heyneighbor · 2026-09-17
- Qwen Code Desktop v0.24.0 ships Linux bwrap sandbox, DingTalk support, cross-session messaging — github-actions[bot] · 2026-09-17
- Nvidia's Agora: 13 LLM Agents Run 12 Days Unsupervised via Git-Based Shared Memory — nvidia · 2026-09-17
- Open-source AgenC ships its biggest release: 658 commits in 28 days, plus a desktop app — tetsuoai · 2026-09-17
- GitHub rewrote the Copilot agent runtime into 800k lines of Rust, mostly written by AI agents — PaulShellDev · 2026-09-17