Are you using Ollama? If so, the issue may be Ollama's default context length: just 2,048 tokens. Ollama truncates the rest of the context silently, so "thinkin

Are you using Ollama? If so, the issue may be Ollama's default context length: just 2,048 tokens. Ollama truncates the rest of the context silently, so "thinking" models cannot work with the default settings. If you are using Ollama, try explicitly setting the `num_ctx` parameter in your request to something higher like 16k or 32k, and then see if you still encounter the looping. I haven't run into that behavior once with this model.