banteg: compaction is now 3x slower — median 1min to 3min across versions
banteg · x · 2026-10-04
banteg observed that compaction latency has degraded 3x across recent versions: median went from 1 min on 5.6 sol to 2.8 mins on 6 astra to 3 mins on 6.1 sol, with p95 at 1.4, 4.3 and 4.8 mins respectively. The same pattern appears on the fast tier. He shared a detailed latency chart.
Related event: Compaction latency degrades across model versions, banteg reports(2 posts)→
More from coding & agent
- AI Workflows Are Only as Reliable as the Way They Fail, Not Their Happy Path — alifcoder · 2026-10-04
- Agent memory is the hidden bottleneck: context management matters as much as reasoning — alifcoder · 2026-10-04
- Analyzing 50K real requests: data collection tops what users want AI agents to automate — quarkcarbon · 2026-10-04
- Lusha launches MCP server letting AI agents do person and company lookups — modelcontextprotocol · 2026-10-04
- Suishijian ships read-only MCP for Chinese calendar, classics and chart-building, seeking feedback — Cassian_Wu · 2026-10-04
- Self-Driving Beginner Project Publishes Roadmap to Learn Full AV Stack in One Simulator — 4310sy · 2026-10-04