High impact

Yes, Claude can do nine loops

What happened

Anthropic demonstrated that its Claude language model can reliably execute up to nine iterative reasoning steps—called “loops”—when solving complex problems, showing that the model can maintain context and improve answers across multiple passes. The researchers achieved this by fine‑tuning Claude with a loop‑aware training regime and testing it on tasks that normally require multi‑step thinking, such as math puzzles and logical deductions. This proves that large language models can be made to think more like humans, revisiting and refining their own outputs without external prompting.

Why it matters

The ability to run multiple internal reasoning loops makes AI assistants more accurate and trustworthy for tasks that need deep, step‑by‑step analysis, potentially reducing the need for human oversight. It also sets a new benchmark for iterative reasoning, pushing the industry toward building models that can self‑improve their answers in real time.

More from Anthropic

Anthropic Upgrades Claude Code Projects to Run Parallel AI Workers Hands-FreeAnthropic · 14 Nov 2026Claude designed FPGA scaler board for 6th-gen consolesAnthropic · 13 Nov 2026I asked Claude Fable 5.2 to turn my site into a 90s Silicon Graphics workstationAnthropic · 13 Nov 2026Anthropic is endorsing SB 53Anthropic · 13 Nov 2026Understanding and addressing AI harmsAnthropic · 13 Nov 2026

This page was collected from anthropic.com and summarised automatically by AIPulsee's pipeline. See how it runs.