Opus 5.5 Keeps Thinking Always On, Defaults to Medium, Uses Compaction for Long Agents
ClaudeDevs · x · 2026-09-23
Anthropic detailed Opus 5.5's reasoning and context management: thinking is always on, Medium is the default, and users can adjust effort.
For long-running agents, compaction summarizes older turns while keeping recent ones intact, keeping a long conversation or agent task inside the context window and the active context small, since response quality degrades as conversations grow.
For more speed, fast mode goes up to 2.5x at $8/$40. The linked docs explain that compaction has Claude write the summary server-side, so no summarization code is needed; both on-demand and token-threshold compaction are offered, with on-demand covering more common use cases, and kept turns retain their thinking on models with preserved thinking.
Related event: Anthropic Launches Claude Opus 5.5: Stronger, 40% Cheaper, Tops AA Index(259 posts)→
More from coding & agent
- lattedit: free local-first LaTeX editor in the browser with a JSON API for agents — cneuralnetwork · 2026-11-09
- Perplexity Computer gains Cloudflare Security Audit skill that traces vulns entry-to-sink — jeffgrimes9 · 2026-11-09
- "Do the needful": the Indian English phrase that works surprisingly well on AI agents — moultano · 2026-11-09
- Self-described AI hater: Claude Code fixed a year-old problem in under an hour — alejandroll10 · 2026-11-09
- Three.js 'Hollow' demo lands as day 7 of a vibe-coding creative streak — thespite · 2026-11-09
- DeepSeek says its Harness aims to be the OS of the agent era, with built-ins replaceable by plugins — dotey · 2026-11-09