Adaptive Thinking

Anthropic’s dynamic reasoning budget allocation introduced 2026-02-09. The model decides per-turn whether to allocate reasoning tokens based on perceived task complexity, rather than applying a fixed budget. Replaces the binary extended-thinking toggle on Claude Opus 4.7.

Key points

• Introduced silently on 2026-02-09 for Opus 4.6 — the change that caused the widely reported “Opus 4.6 degradation” ^[010]
• When introduced, the default effort was set to “medium”, causing a 73% collapse in reasoning depth (2,200 → 600 characters of reasoning per turn) and 33.7% of sessions skipping file reads ^[010]
• Boris Cherny (Claude Code creator) confirmed: turns with zero reasoning had the highest hallucination rates; turns with deep reasoning were correct ^[010]
• Opus 4.7 uses adaptive thinking with a higher default and the new “X high” ceiling — Anthropic’s stated cure for the 4.6 degradation they caused ^[010]
• Community analysis of 18K thinking blocks across 7K sessions confirmed: thinking depth dropped 67% and “edit without reading” behaviour rose from 6% to 34% during the Opus 4.6 degradation period ^[011]

Related entities

• Anthropic — the developer who introduced this change
• Claude Opus 4 7 — first model to replace extended thinking toggle with adaptive thinking

Related concepts

• Effort Levels — the hierarchy of reasoning budgets adaptive thinking selects from
• Context Rot — low-effort (zero-reasoning) turns are a primary cause of context rot symptoms
• Token Economics — adaptive thinking directly controls per-turn token consumption

Source references

• ^[010] Nate Herk — Cloud agents & model releases cluster (2026-04-14 to 2026-04-17)
• ^[011] Nate Herk — Claude Code power features cluster (2026-04-20 to 2026-04-27)