I saw something about the Todos in the change notes, but it’s not clear how that works. Is there a link to docs? Or can someone explain? Tx
Hi X here, avid LLM user.
You must use the todo_tool (explicitly request the LLM to use the tool). I only use Claude 3.7 Thinking for clarification.
Example: I’m making a project that “insert your projects overview”, Use the todo_tool in order to create a full work flow from start to finish based on severity and critical implementations. Stay concise and comprehensive.
If it works, you will be greeted with this:
EDIT: Ensure you have it enabled first 
why do you use 3.7 thinking? how better is it compared to 4?
1 Like
This is only my experience regarding my use of working with all LLMs over the course of 5+ months spending over $2000 using even the most powerful and the daily drivers etc.
I originally was using Auto for the longest until I discovered the description “Uses the best premium model based on performance and speed”, this led to me discover while it may explicitly state these functions, I was only ever given ChatGPT. I can tell because I have learned precisely how each LLM talks and communicated both front-end and back-end. The obvious indicator being the excessive emojis and “joyful” take on everything and the overall lack of professionalism or “industry level”.
My experience with Claude started amazing with 3.5 and was affordable all while remaining intelligent. Since this new update with the updated rates I have only experienced errors and failures (LLM being non-compliant) or just straight brain dead?
My point being, Claude 3.7 works great but I’m making a full game engine based on LITHIUM and I require more “layers” of processing power and logical thinking so I have to use thinking. 3.7 only cost $0.08 vs $0.04 for other similar models like Sonnet 3.5 for example.
Any other LLM such as Opus or o3-pro is to expensive and I can’t afford to spend the extra $ so I have optimized a cheaper rate with a cheaper LLM with the most advanced capabilities required for my work flow.
To finish this off, Claude 4 = CHATGPT. Is it a bug? Is it just me? I have zero idea but I can’t use it AT ALL.
Edit: To answer your question – On any given day when BOTH LLM’s are working to the upmost functionality, Sonnet 4 without question wins. In the circumstances where the IDE is causing bad functionality, 3.7 THINKING wins (Just my experience again)
2 Likes
thank you!
I have the To-Do List feature enabled and explicitly asked for todo_tool to be used. Sonnet 4 is claiming that tool doesn’t exist. Is this tool only available in a fresh workspace? Auto and Sonnet 4 are showing only the following list of tools:
- codebase_search
- read_file
- list_dir
- grep_search
- file_search
Could I help myself with your vote? To-do lists don't work