Most modern LLMs (especially frontier ones) are large token hogs because they draft, check, re-draft, the content (whether an output message; or a code diff) sometimes multiple times in the thinking block.

When you see a thinking summary like "Now writing the function..."; the raw thinking is actually writing the function in its internal thinking. Occasionally, the summariser misses and you get to see the raw text from models like Opus.

You can also try an open weight LLM like Qwen3.6 and see something that probably resembles the shape of frontier model thinking in some loose way.