Claude Code 2.1.296 regression: auto-compaction fires at 281K–845K tokens vs documented ~967K
BBJeanz · ghdev · 2026-10-10
Claude Code issue #101067 reports a regression in 2.1.296: auto-compaction on claude-opus-5-5 fires far below the documented default.
- Expected: models with native 1M window compact at 967K tokens
- Actual: versions ≤2.1.294 compacted 47/47 times at 966K–981K; 2.1.296 compacted 8 times scattered across 281K–845K, with no fixed threshold (one session hit 283K, 532K and 425K)
- Config ruled out: no autoCompactWindow, related env vars, or launch flags
- Impact: each compaction drops working context and takes 50–83s, so long sessions compact 3x more often
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23