llm-echo 0.5 released, simulated token counting now includes tool calls and reasoning tokens

llm-echo is a debugging plugin maintained by Simon Willison that provides a "model" without involving a real LLM, simply echoing back the prompt it receives. Version 0.5 updates the simulated token usage count to include tool calls, system prompts, and reasoning tokens. The example in the post uses uvx --with llm-echo with llm -m echo, enabling thinking and --usage, and the output shows Token usage: 2 input, 24 output. The real impact is for developers building LLM plugins and debugging prompt flows: previously the simulated count only estimated simple input and output, so numbers became inaccurate once tool calls or reasoning tokens were involved; now, before running against a real model, you can use the echo model to verify usage display and cost estimates.