Qwen-32B

Local LLMs Lose Tool Call Accuracy After 8–9 Chained Calls

Qwen 32B, Gemma 9B, and Command R 32B all fail similarly after 8+ tool calls — attention dilution, not context limits.